jyapayne/langflow
1
0
Fork 0
Code Issues Pull requests Projects Releases 1 Packages Wiki Activity Actions

Merge dev into types_refactor

Browse source
This commit is contained in:
Igor Carvalho 2023-08-02 14:30:10 -03:00
commit a4319eaa43
149 changed files with 9170 additions and 2704 deletions
Download patch file Download diff file Expand all files Collapse all files

2
.gitignore vendored
View file

@ -251,3 +251,5 @@ langflow.db
# docusaurus
.docusaurus/
/tmp/*

3
.vscode/launch.json vendored
View file

@ -6,7 +6,8 @@
"request": "launch",
"module": "uvicorn",
"args": [
"langflow.main:app",
"--factory",
"langflow.main:create_app",
"--port",
"7860",
"--reload",

41
CONTRIBUTING.md
View file

@ -9,7 +9,7 @@ Please do not try to push directly to this repo unless you are a maintainer.
## 🗺️Contributing Guidelines
### 🚩GitHub Issues
## 🚩GitHub Issues
Our [issues](https://github.com/logspace-ai/langflow/issues) page is kept up to date
with bugs, improvements, and feature requests. There is a taxonomy of labels to help
@ -33,18 +33,19 @@ so that more people can benefit from it.
[collapses the content](https://developer.mozilla.org/en/docs/Web/HTML/Element/details)
so it only becomes visible on click, making the issue easier to read and follow.
### Issue labels
## Issue labels
[See this page](https://github.com/logspace-ai/langflow/labels) for an overview of
the system we use to tag our issues and pull requests.
## Local development
### Local development
You can develop Langflow using docker compose, or locally.
We provide a .vscode/launch.json file for debugging the backend in VSCode, which is a lot faster than using docker compose.
Setting up hooks:
```bash
make init
```
@ -53,30 +54,46 @@ This will install the pre-commit hooks, which will run `make format` on every co
It is advised to run `make lint` before pushing to the repository.
#### **Locally**
Run locally by cloning the repository and installing the dependencies. We recommend using a virtual environment to isolate the dependencies from your system.
## Run locally
Langflow can run locally by cloning the repository and installing the dependencies. We recommend using a virtual environment to isolate the dependencies from your system.
Before you start, make sure you have the following installed:
- Poetry (>=1.4)
- Node.js
For the backend, you will need to install the dependencies and start the development server.
- Poetry (>=1.4)
- Node.js
Then, in the root folder, install the dependencies and start the development server for the backend:
```bash
make install_backend
make backend
```
For the frontend, you will need to install the dependencies and start the development server.
And the frontend:
```bash
make frontend
```
## Docker compose
The following snippet will run the backend and frontend in separate containers. The frontend will be available at `localhost:3000` and the backend at `localhost:7860`.
#### **Docker compose**
This will run the backend and frontend in separate containers. The frontend will be available at `localhost:3000` and the backend at `localhost:7860`.
```bash
docker compose up --build
# or
make dev build=1
```
## Documentation
The documentation is built using [Docusaurus](https://docusaurus.io/). To run the documentation locally, run the following commands:
```bash
cd docs
npm install
npm run start
```
The documentation will be available at `localhost:3000` and all the files are located in the `docs/docs` folder.
Once you are done with your changes, you can create a Pull Request to the `main` branch.

2
Makefile
View file

@ -46,7 +46,7 @@ install_backend:
backend:
make install_backend
poetry run uvicorn src.backend.langflow.main:app --port 7860 --reload --log-level debug
poetry run uvicorn --factory src.backend.langflow.main:create_app --port 7860 --reload --log-level debug
build_and_run:
echo 'Removing dist folder'

143
README.md
View file

@ -13,7 +13,6 @@
<img alt="Github License" src="https://img.shields.io/github/license/logspace-ai/langflow" />
</p>
<p>
<a href="https://discord.gg/EqksyE2EX9"><img alt="Discord Server" src="https://dcbadge.vercel.app/api/server/EqksyE2EX9?compact=true&style=flat"/></a>
<a href="https://huggingface.co/spaces/Logspace/Langflow"><img src="https://huggingface.co/datasets/huggingface/badges/raw/main/open-in-hf-spaces-sm.svg" alt="HuggingFace Spaces"></a>
@ -22,54 +21,75 @@
<a href="https://github.com/logspace-ai/langflow">
<img width="100%" src="https://github.com/logspace-ai/langflow/blob/main/img/langflow-demo.gif?raw=true"></a>
<p>
</p>
# Table of Contents
- [⛓️ Langflow](#️-langflow)
- [Table of Contents](#table-of-contents)
- [📦 Installation](#-installation)
- [Locally](#locally)
- [HuggingFace Spaces](#huggingface-spaces)
- [Locally](#locally)
- [HuggingFace Spaces](#huggingface-spaces)
- [🖥️ Command Line Interface (CLI)](#️-command-line-interface-cli)
- [Usage](#usage)
- [Environment Variables](#environment-variables)
- [Usage](#usage)
- [Environment Variables](#environment-variables)
- [Deployment](#deployment)
- [Deploy Langflow on Google Cloud Platform](#deploy-langflow-on-google-cloud-platform)
- [Deploy Langflow on Jina AI Cloud](#deploy-langflow-on-jina-ai-cloud)
- [API Usage](#api-usage)
- [API Usage](#api-usage)
- [Deploy on Railway](#deploy-on-railway)
- [Deploy on Render](#deploy-on-render)
- [🎨 Creating Flows](#-creating-flows)
- [👋 Contributing](#-contributing)
- [📄 License](#-license)
# 📦 Installation
### <b>Locally</b>
You can install Langflow from pip:
```shell
# This installs the package without dependencies for local models
pip install langflow
```
To use local models (e.g llama-cpp-python) run:
```shell
pip install langflow[local]
```
This will install the following dependencies:
- [CTransformers](https://github.com/marella/ctransformers)
- [llama-cpp-python](https://github.com/abetlen/llama-cpp-python)
- [sentence-transformers](https://github.com/UKPLab/sentence-transformers)
You can still use models from projects like LocalAI
Next, run:
```shell
python -m langflow
```
or
```shell
langflow # or langflow --help
```
### HuggingFace Spaces
You can also check it out on [HuggingFace Spaces](https://huggingface.co/spaces/Logspace/Langflow) and run it in your browser! You can even clone it and have your own copy of Langflow to play with.
# 🖥️ Command Line Interface (CLI)
Langflow provides a command-line interface (CLI) for easy management and configuration.
### Usage
## Usage
You can run the Langflow using the following command:
@ -87,6 +107,7 @@ Each option is detailed below:
- `--config`: Defines the path to the configuration file. The default is `config.yaml`.
- `--env-file`: Specifies the path to the .env file containing environment variables. The default is `.env`.
- `--log-level`: Defines the logging level. Can be set using the `LANGFLOW_LOG_LEVEL` environment variable. The default is `critical`.
- `--components-path`: Specifies the path to the directory containing custom components. Can be set using the `LANGFLOW_COMPONENTS_PATH` environment variable. The default is `langflow/components`.
- `--log-file`: Specifies the path to the log file. Can be set using the `LANGFLOW_LOG_FILE` environment variable. The default is `logs/langflow.log`.
- `--cache`: Selects the type of cache to use. Options are `InMemoryCache` and `SQLiteCache`. Can be set using the `LANGFLOW_LANGCHAIN_CACHE` environment variable. The default is `SQLiteCache`.
- `--jcloud/--no-jcloud`: Toggles the option to deploy on Jina AI Cloud. The default is `no-jcloud`.
@ -114,7 +135,6 @@ Alternatively, click the **"Open in Cloud Shell"** button below to launch Google
[![Open in Cloud Shell](https://gstatic.com/cloudssh/images/open-btn.svg)](https://console.cloud.google.com/cloudshell/open?git_repo=https://github.com/logspace-ai/langflow&working_dir=scripts&shellonly=true&tutorial=walkthroughtutorial_spot.md)
## Deploy Langflow on [Jina AI Cloud](https://github.com/jina-ai/langchain-serve)
Langflow integrates with langchain-serve to provide a one-command deployment to Jina AI Cloud.
@ -122,6 +142,8 @@ Langflow integrates with langchain-serve to provide a one-command deployment to
Start by installing `langchain-serve` with
```bash
pip install langflow[deploy]
# or
pip install -U langchain-serve
```
@ -140,33 +162,33 @@ langflow --jcloud
<details>
<summary>Show complete (example) output</summary>
```text
🚀 Deploying Langflow server on Jina AI Cloud
╭───────────────────────── 🎉 Flow is available! ──────────────────────────╮
│ │
│ ID langflow-e3dd8820ec │
│ Gateway (Websocket) wss://langflow-e3dd8820ec.wolf.jina.ai │
│ Dashboard https://dashboard.wolf.jina.ai/flow/e3dd8820ec │
│ │
╰──────────────────────────────────────────────────────────────────────────╯
╭──────────────┬──────────────────────────────────────────────────────────────────────────────╮
│ App ID │ langflow-e3dd8820ec │
├──────────────┼──────────────────────────────────────────────────────────────────────────────┤
│ Phase │ Serving │
├──────────────┼──────────────────────────────────────────────────────────────────────────────┤
│ Endpoint │ wss://langflow-e3dd8820ec.wolf.jina.ai │
├──────────────┼──────────────────────────────────────────────────────────────────────────────┤
│ App logs │ dashboards.wolf.jina.ai │
├──────────────┼──────────────────────────────────────────────────────────────────────────────┤
│ Swagger UI │ https://langflow-e3dd8820ec.wolf.jina.ai/docs │
├──────────────┼──────────────────────────────────────────────────────────────────────────────┤
│ OpenAPI JSON │ https://langflow-e3dd8820ec.wolf.jina.ai/openapi.json │
╰──────────────┴──────────────────────────────────────────────────────────────────────────────╯
```text
🚀 Deploying Langflow server on Jina AI Cloud
╭───────────────────────── 🎉 Flow is available! ──────────────────────────╮
│ │
│ ID langflow-e3dd8820ec │
│ Gateway (Websocket) wss://langflow-e3dd8820ec.wolf.jina.ai │
│ Dashboard https://dashboard.wolf.jina.ai/flow/e3dd8820ec │
│ │
╰──────────────────────────────────────────────────────────────────────────╯
╭──────────────┬──────────────────────────────────────────────────────────────────────────────╮
│ App ID │ langflow-e3dd8820ec │
├──────────────┼──────────────────────────────────────────────────────────────────────────────┤
│ Phase │ Serving │
├──────────────┼──────────────────────────────────────────────────────────────────────────────┤
│ Endpoint │ wss://langflow-e3dd8820ec.wolf.jina.ai │
├──────────────┼──────────────────────────────────────────────────────────────────────────────┤
│ App logs │ dashboards.wolf.jina.ai │
├──────────────┼──────────────────────────────────────────────────────────────────────────────┤
│ Swagger UI │ https://langflow-e3dd8820ec.wolf.jina.ai/docs │
├──────────────┼──────────────────────────────────────────────────────────────────────────────┤
│ OpenAPI JSON │ https://langflow-e3dd8820ec.wolf.jina.ai/openapi.json │
╰──────────────┴──────────────────────────────────────────────────────────────────────────────╯
🎉 Langflow server successfully deployed on Jina AI Cloud 🎉
🔗 Click on the link to open the server (please allow ~1-2 minutes for the server to startup): https://langflow-e3dd8820ec.wolf.jina.ai/
📖 Read more about managing the server: https://github.com/jina-ai/langchain-serve
```
🎉 Langflow server successfully deployed on Jina AI Cloud 🎉
🔗 Click on the link to open the server (please allow ~1-2 minutes for the server to startup): https://langflow-e3dd8820ec.wolf.jina.ai/
📖 Read more about managing the server: https://github.com/jina-ai/langchain-serve
```
</details>
@ -177,7 +199,7 @@ You can use Langflow directly on your browser, or use the API endpoints on Jina
<details>
<summary>Show API usage (with python)</summary>
```python
```python
import requests
BASE_API_URL = "https://langflow-e3dd8820ec.wolf.jina.ai/api/v1/predict"
@ -185,47 +207,49 @@ FLOW_ID = "864c4f98-2e59-468b-8e13-79cd8da07468"
# You can tweak the flow by adding a tweaks dictionary
# e.g {"OpenAI-XXXXX": {"model_name": "gpt-4"}}
TWEAKS = {
"ChatOpenAI-g4jEr": {},
"ConversationChain-UidfJ": {}
"ChatOpenAI-g4jEr": {},
"ConversationChain-UidfJ": {}
}
def run_flow(message: str, flow_id: str, tweaks: dict = None) -> dict:
"""
Run a flow with a given message and optional tweaks.
"""
Run a flow with a given message and optional tweaks.
:param message: The message to send to the flow
:param flow_id: The ID of the flow to run
:param tweaks: Optional tweaks to customize the flow
:return: The JSON response from the flow
"""
api_url = f"{BASE_API_URL}/{flow_id}"
:param message: The message to send to the flow
:param flow_id: The ID of the flow to run
:param tweaks: Optional tweaks to customize the flow
:return: The JSON response from the flow
"""
api_url = f"{BASE_API_URL}/{flow_id}"
payload = {"message": message}
payload = {"message": message}
if tweaks:
payload["tweaks"] = tweaks
if tweaks:
payload["tweaks"] = tweaks
response = requests.post(api_url, json=payload)
return response.json()
response = requests.post(api_url, json=payload)
return response.json()
# Setup any tweaks you want to apply to the flow
print(run_flow("Your message", flow_id=FLOW_ID, tweaks=TWEAKS))
```
```
```json
{
"result": "Great choice! Bangalore in the 1920s was a vibrant city with a rich cultural and political scene. Here are some suggestions for things to see and do:\n\n1. Visit the Bangalore Palace - built in 1887, this stunning palace is a perfect example of Tudor-style architecture. It was home to the Maharaja of Mysore and is now open to the public.\n\n2. Attend a performance at the Ravindra Kalakshetra - this cultural center was built in the 1920s and is still a popular venue for music and dance performances.\n\n3. Explore the neighborhoods of Basavanagudi and Malleswaram - both of these areas have retained much of their old-world charm and are great places to walk around and soak up the atmosphere.\n\n4. Check out the Bangalore Club - founded in 1868, this exclusive social club was a favorite haunt of the British expat community in the 1920s.\n\n5. Attend a meeting of the Indian National Congress - founded in 1885, the INC was a major force in the Indian independence movement and held many meetings and rallies in Bangalore in the 1920s.\n\nHope you enjoy your trip to 1920s Bangalore!"
}
```
```json
{
"result": "Great choice! Bangalore in the 1920s was a vibrant city with a rich cultural and political scene. Here are some suggestions for things to see and do:\n\n1. Visit the Bangalore Palace - built in 1887, this stunning palace is a perfect example of Tudor-style architecture. It was home to the Maharaja of Mysore and is now open to the public.\n\n2. Attend a performance at the Ravindra Kalakshetra - this cultural center was built in the 1920s and is still a popular venue for music and dance performances.\n\n3. Explore the neighborhoods of Basavanagudi and Malleswaram - both of these areas have retained much of their old-world charm and are great places to walk around and soak up the atmosphere.\n\n4. Check out the Bangalore Club - founded in 1868, this exclusive social club was a favorite haunt of the British expat community in the 1920s.\n\n5. Attend a meeting of the Indian National Congress - founded in 1885, the INC was a major force in the Indian independence movement and held many meetings and rallies in Bangalore in the 1920s.\n\nHope you enjoy your trip to 1920s Bangalore!"
}
```
</details>
> Read more about resource customization, cost, and management of Langflow apps on Jina AI Cloud in the **[langchain-serve](https://github.com/jina-ai/langchain-serve)** repository.
## Deploy on Railway
[![Deploy on Railway](https://railway.app/button.svg)](https://railway.app/template/Emy2sU?referralCode=MnPSdg)
## Deploy on Render
<a href="https://render.com/deploy?repo=https://github.com/logspace-ai/langflow/tree/main">
<img src="https://render.com/images/deploy-to-render-button.svg" alt="Deploy to Render" />
</a>
@ -248,12 +272,10 @@ flow = load_flow_from_json("path/to/flow.json")
flow("Hey, have you heard of Langflow?")
```
# 👋 Contributing
We welcome contributions from developers of all levels to our open-source project on GitHub. If you'd like to contribute, please check our [contributing guidelines](./CONTRIBUTING.md) and help make Langflow more accessible.
Join our [Discord](https://discord.com/invite/EqksyE2EX9) server to ask questions, make suggestions and showcase your projects! 🦾
<p>
@ -261,7 +283,6 @@ Join our [Discord](https://discord.com/invite/EqksyE2EX9) server to ask question
[![Star History Chart](https://api.star-history.com/svg?repos=logspace-ai/langflow&type=Timeline)](https://star-history.com/#logspace-ai/langflow&Date)
# 📄 License
Langflow is released under the MIT License. See the LICENSE file for details.

9
docs/docs/components/agents.mdx
View file

@ -1,5 +1,14 @@
import Admonition from '@theme/Admonition';
# Agents
<Admonition type="caution" icon="🚧" title="ZONE UNDER CONSTRUCTION">
<p>
We appreciate your understanding as we polish our documentation – it may contain some rough edges. Share your feedback or report issues to help us improve! 🛠️📝
</p>
</Admonition>
Agents are components that use reasoning to make decisions and take actions, designed to autonomously perform tasks or provide services with some degree of “freedom” (or agency). They combine the power of LLM chaining processes with access to external tools such as APIs to interact with applications and accomplish tasks.
---

49
docs/docs/components/chains.mdx
View file

@ -1,9 +1,18 @@
import ThemedImage from "@theme/ThemedImage";
import useBaseUrl from "@docusaurus/useBaseUrl";
import ZoomableImage from "/src/theme/ZoomableImage.js";
import Admonition from "@theme/Admonition";
# Chains
<Admonition type="caution" icon="🚧" title="ZONE UNDER CONSTRUCTION">
<p>
We appreciate your understanding as we polish our documentation – it may
contain some rough edges. Share your feedback or report issues to help us
improve! 🛠️📝
</p>
</Admonition>
Chains, in the context of language models, refer to a series of calls made to a language model. It allows for the output of one call to be used as the input for another call. Different types of chains allow for different levels of complexity. Chains are useful for creating pipelines and executing specific scenarios.
---
@ -12,22 +21,23 @@ Chains, in the context of language models, refer to a series of calls made to a
The `CombineDocsChain` incorporates methods to combine or aggregate loaded documents for question-answering functionality.
:::info
<Admonition type="info">
Works as a proxy of LangChain’s [documents](https://python.langchain.com/docs/modules/chains/document/) chains generated by the `load_qa_chain` function.
:::
</Admonition>
**Params**
- **LLM:** Language Model to use in the chain.
- **chain_type:** The chain type to be used. Each one of them applies a different “combination strategy”.
- **stuff**: The stuff [documents](https://python.langchain.com/docs/modules/chains/document/stuff) chain (“stuff" as in "to stuff" or "to fill") is the most straightforward of *the* document chains. It takes a list of documents, inserts them all into a prompt, and passes that prompt to an LLM. This chain is well-suited for applications where documents are small and only a few are passed in for most calls.
- **map_reduce**: The map-reduce [documents](https://python.langchain.com/docs/modules/chains/document/map_reduce) chain first applies an LLM chain to each document individually (the Map step), treating the chain output as a new document. It then passes all the new documents to a separate combined documents chain to get a single output (the Reduce step). It can optionally first compress or collapse the mapped documents to make sure that they fit in the combined documents chain (which will often pass them to an LLM). This compression step is performed recursively if necessary.
- **map_rerank**: The map re-rank [documents](https://python.langchain.com/docs/modules/chains/document/map_rerank) chain runs an initial prompt on each document that not only tries to complete a task but also gives a score for how certain it is in its answer. The highest-scoring response is returned.
- **refine**: The refine [documents](https://python.langchain.com/docs/modules/chains/document/refine) chain constructs a response by looping over the input documents and iteratively updating its answer. For each document, it passes all non-document inputs, the current document, and the latest intermediate answer to an LLM chain to get a new answer.
Since the Refine chain only passes a single document to the LLM at a time, it is well-suited for tasks that require analyzing more documents than can fit in the model's context. The obvious tradeoff is that this chain will make far more LLM calls than, for example, the Stuff documents chain. There are also certain tasks that are difficult to accomplish iteratively. For example, the Refine chain can perform poorly when documents frequently cross-reference one another or when a task requires detailed information from many documents.
- **stuff**: The stuff [documents](https://python.langchain.com/docs/modules/chains/document/stuff) chain (“stuff" as in "to stuff" or "to fill") is the most straightforward of _the_ document chains. It takes a list of documents, inserts them all into a prompt, and passes that prompt to an LLM. This chain is well-suited for applications where documents are small and only a few are passed in for most calls.
- **map_reduce**: The map-reduce [documents](https://python.langchain.com/docs/modules/chains/document/map_reduce) chain first applies an LLM chain to each document individually (the Map step), treating the chain output as a new document. It then passes all the new documents to a separate combined documents chain to get a single output (the Reduce step). It can optionally first compress or collapse the mapped documents to make sure that they fit in the combined documents chain (which will often pass them to an LLM). This compression step is performed recursively if necessary.
- **map_rerank**: The map re-rank [documents](https://python.langchain.com/docs/modules/chains/document/map_rerank) chain runs an initial prompt on each document that not only tries to complete a task but also gives a score for how certain it is in its answer. The highest-scoring response is returned.
- **refine**: The refine [documents](https://python.langchain.com/docs/modules/chains/document/refine) chain constructs a response by looping over the input documents and iteratively updating its answer. For each document, it passes all non-document inputs, the current document, and the latest intermediate answer to an LLM chain to get a new answer.
Since the Refine chain only passes a single document to the LLM at a time, it is well-suited for tasks that require analyzing more documents than can fit in the model's context. The obvious tradeoff is that this chain will make far more LLM calls than, for example, the Stuff documents chain. There are also certain tasks that are difficult to accomplish iteratively. For example, the Refine chain can perform poorly when documents frequently cross-reference one another or when a task requires detailed information from many documents.
---
@ -41,7 +51,7 @@ The `ConversationChain` is a straightforward chain for interactive conversations
- **Memory:** Default memory store.
- **input_key:** Used to specify the key under which the user input will be stored in the conversation memory. It allows you to provide the user's input to the chain for processing and generating a response.
- **output_key:** Used to specify the key under which the generated response will be stored in the conversation memory. It allows you to retrieve the response using the specified key.
- **verbose:** This parameter is used to control the level of detail in the output of the chain. When set to True, it will print out some internal states of the chain while it is being run, which can be helpful for debugging and understanding the chain's behavior. If set to False, it will suppress the verbose output — defaults to `False`.
- **verbose:** This parameter is used to control the level of detail in the output of the chain. When set to True, it will print out some internal states of the chain while it is being run, which can be helpful for debugging and understanding the chain's behavior. If set to False, it will suppress the verbose output — defaults to `False`.
---
@ -49,11 +59,11 @@ The `ConversationChain` is a straightforward chain for interactive conversations
The `ConversationalRetrievalChain` extracts information and provides answers by combining document search and question-answering abilities.
:::info
<Admonition type="info">
A retriever is a component that finds documents based on a query. It doesn't store the documents themselves, but it returns the ones that match the query.
:::
</Admonition >
**Params**
@ -61,12 +71,13 @@ A retriever is a component that finds documents based on a query. It doesn't sto
- **Memory:** Default memory store.
- **Retriever:** The retriever used to fetch relevant documents.
- **chain_type:** The chain type to be used. Each one of them applies a different “combination strategy”.
- **stuff**: The stuff [documents](https://python.langchain.com/docs/modules/chains/document/stuff) chain (“stuff" as in "to stuff" or "to fill") is the most straightforward of *the* document chains. It takes a list of documents, inserts them all into a prompt, and passes that prompt to an LLM. This chain is well-suited for applications where documents are small and only a few are passed in for most calls.
- **map_reduce**: The map-reduce [documents](https://python.langchain.com/docs/modules/chains/document/map_reduce) chain first applies an LLM chain to each document individually (the Map step), treating the chain output as a new document. It then passes all the new documents to a separate combined documents chain to get a single output (the Reduce step). It can optionally first compress or collapse the mapped documents to make sure that they fit in the combined documents chain (which will often pass them to an LLM). This compression step is performed recursively if necessary.
- **map_rerank**: The map re-rank [documents](https://python.langchain.com/docs/modules/chains/document/map_rerank) chain runs an initial prompt on each document that not only tries to complete a task but also gives a score for how certain it is in its answer. The highest-scoring response is returned.
- **refine**: The refine [documents](https://python.langchain.com/docs/modules/chains/document/refine) chain constructs a response by looping over the input documents and iteratively updating its answer. For each document, it passes all non-document inputs, the current document, and the latest intermediate answer to an LLM chain to get a new answer.
Since the Refine chain only passes a single document to the LLM at a time, it is well-suited for tasks that require analyzing more documents than can fit in the model's context. The obvious tradeoff is that this chain will make far more LLM calls than, for example, the Stuff documents chain. There are also certain tasks that are difficult to accomplish iteratively. For example, the Refine chain can perform poorly when documents frequently cross-reference one another or when a task requires detailed information from many documents.
- **stuff**: The stuff [documents](https://python.langchain.com/docs/modules/chains/document/stuff) chain (“stuff" as in "to stuff" or "to fill") is the most straightforward of _the_ document chains. It takes a list of documents, inserts them all into a prompt, and passes that prompt to an LLM. This chain is well-suited for applications where documents are small and only a few are passed in for most calls.
- **map_reduce**: The map-reduce [documents](https://python.langchain.com/docs/modules/chains/document/map_reduce) chain first applies an LLM chain to each document individually (the Map step), treating the chain output as a new document. It then passes all the new documents to a separate combined documents chain to get a single output (the Reduce step). It can optionally first compress or collapse the mapped documents to make sure that they fit in the combined documents chain (which will often pass them to an LLM). This compression step is performed recursively if necessary.
- **map_rerank**: The map re-rank [documents](https://python.langchain.com/docs/modules/chains/document/map_rerank) chain runs an initial prompt on each document that not only tries to complete a task but also gives a score for how certain it is in its answer. The highest-scoring response is returned.
- **refine**: The refine [documents](https://python.langchain.com/docs/modules/chains/document/refine) chain constructs a response by looping over the input documents and iteratively updating its answer. For each document, it passes all non-document inputs, the current document, and the latest intermediate answer to an LLM chain to get a new answer.
Since the Refine chain only passes a single document to the LLM at a time, it is well-suited for tasks that require analyzing more documents than can fit in the model's context. The obvious tradeoff is that this chain will make far more LLM calls than, for example, the Stuff documents chain. There are also certain tasks that are difficult to accomplish iteratively. For example, the Refine chain can perform poorly when documents frequently cross-reference one another or when a task requires detailed information from many documents.
- **return_source_documents:** Used to specify whether or not to include the source documents that were used to answer the question in the output. When set to `True`, source documents will be included in the output along with the generated answer. This can be useful for providing additional context or references to the user — defaults to `True`.
- **verbose:** Whether or not to run in verbose mode. In verbose mode, intermediate logs will be printed to the console — defaults to `False`.
@ -108,17 +119,17 @@ The `LLMMathChain` works by using the language model with an `LLMChain` to under
`RetrievalQA` is a chain used to find relevant documents or information to answer a given query. The retriever is responsible for returning the relevant documents based on the query, and the QA component then extracts the answer from those documents. The retrieval QA system combines the capabilities of both the retriever and the QA component to provide accurate and relevant answers to user queries.
:::info
<Admonition type="info">
A retriever is a component that finds documents based on a query. It doesn't store the documents themselves, but it returns the ones that match the query.
:::
</Admonition >
**Params**
- **Combine Documents Chain:** Chain to use to combine the documents.
- **Memory:** Default memory store.
- **Retriever:** The retriever used to fetch relevant documents.
- **Retriever:** The retriever used to fetch relevant documents.
- **input_key:** This parameter is used to specify the key in the input data that contains the question. It is used to retrieve the question from the input data and pass it to the question-answering model for generating the answer — defaults to `query`.
- **output_key:** This parameter is used to specify the key in the output data where the generated answer will be stored. It is used to retrieve the answer from the output data after the question-answering model has generated it — defaults to `result`.
- **return_source_documents:** Used to specify whether or not to include the source documents that were used to answer the question in the output. When set to `True`, source documents will be included in the output along with the generated answer. This can be useful for providing additional context or references to the user — defaults to `True`.
@ -134,4 +145,4 @@ The `SQLDatabaseChain` finds answers to questions using a SQL database. It works
- **Db:** SQL Database to connect to.
- **LLM:** Language Model to use in the chain.
- **Prompt:** Prompt template to translate natural language to SQL.
- **Prompt:** Prompt template to translate natural language to SQL.

92
docs/docs/components/custom.mdx Normal file
View file

@ -0,0 +1,92 @@
import Admonition from "@theme/Admonition";
# Custom Components
Used to create a custom component, a special type of Langflow component that allows users to extend the functionality of the platform by creating their own reusable and configurable components from a Python script.
To use a custom component, follow these steps:
- Create a class that inherits from _`langflow.CustomComponent`_ and contains a _`build`_ method.
- Use arguments with [Type Annotations (or Type Hints)](https://docs.python.org/3/library/typing.html) of the _`build`_ method to create component fields.
- If applicable, use the _`build_config`_ method to customize how these fields look and behave.
<Admonition type="info" label="Tip">
For an in-depth explanation of custom components, their rules, and applications, make sure to read [Custom Component guidelines](../guidelines/custom-component).
</Admonition>
**Params**
- **Code:** The Python code to define the component.
## The CustomComponent Class
The CustomComponent class serves as the foundation for creating custom components. By inheriting this class, users can create new, configurable components, tailored to their specific requirements.
**Methods**
- **build**: This method is required within a Custom Component class. It defines the component's functionality and specifies how it processes input data to produce output data. This method is called when the component is built (i.e., when you click the _Build_ ⚡ button in the canvas).
The type annotations of the _`build`_ instance method are used to create the fields of the component.
| Supported Types |
| --------------------------------------------------------- |
| _`str`_, _`int`_, _`float`_, _`bool`_, _`list`_, _`dict`_ |
| _`langchain.chains.base.Chain`_ |
| _`langchain.PromptTemplate`_ |
| _`langchain.llms.base.BaseLLM`_ |
| _`langchain.Tool`_ |
| _`langchain.document_loaders.base.BaseLoader`_ |
| _`langchain.schema.Document`_ |
| _`langchain.text_splitters.TextSplitter`_ |
| _`langchain.vectorstores.base.VectorStore`_ |
| _`langchain.embeddings.base.Embeddings`_ |
| _`langchain.schema.BaseRetriever`_ |
<Admonition type="info">
Unlike Langchain types, base Python types do not add a
[handle](../guidelines/components) to the field by default. To add handles,
use the _`input_types`_ key in the _`build_config`_ method.
</Admonition>
- **build_config**: Used to define the configuration fields of the component (if applicable). It should always return a dictionary with specific keys representing the field names and corresponding configurations. This method is called when the code is processed (i.e., when you click _Check and Save_ in the code editor). It must follow the format described below:
- Top-level keys are field names.
- Their values are also of type _`dict`_. They specify the behavior of the generated fields.
Below are the available keys used to configure component fields:
| Key | Description |
| -------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| _`field_type: str`_ | The type of the field (can be any of the types supported by the _`build`_ method). |
| _`is_list: bool`_ | If the field can be a list of values, meaning that the user can manually add more inputs to the same field. |
| _`options: List[str]`_ | When defined, the field becomes a dropdown menu where a list of strings defines the options to be displayed. If the _`value`_ attribute is set to one of the options, that option becomes default. For this parameter to work, _`field_type`_ should invariably be _`str`_. |
| _`multiline: bool`_ | Defines if a string field opens a text editor. Useful for longer texts. |
| _`input_types: List[str]`_ | Used when you want a _`str`_ field to have connectable handles. |
| _`display_name: str`_ | Defines the name of the field. |
| _`advanced: bool`_ | Hide the field in the canvas view (displayed component settings only). Useful when a field is for advanced users. |
| _`password: bool`_ | To mask the input text. Useful to hide sensitive text (e.g. API keys). |
| _`required: bool`_ | Makes the field required. |
| _`info: str`_ | Adds a tooltip to the field. |
| _`file_types: List[str]`_ | This is a requirement if the _`field_type`_ is _file_. Defines which file types will be accepted. For example, _json_, _yaml_ or _yml_. |
- The CustomComponent class also provides helpful methods for specific tasks (e.g., to load and use other flows from the Langflow platform):
| Method Name | Description |
| -------------- | ------------------------------------------------------------------- |
| _`list_flows`_ | Returns a list of Flow objects with an _`id`_ and a _`name`_. |
| _`get_flow`_ | Returns a Flow object. Parameters are _`flow_name`_ or _`flow_id`_. |
| _`load_flow`_ | Loads a flow from a given _`id`_. |
- Useful attributes:
| Attribute Name | Description |
| -------------- | ----------------------------------------------------------------------------- |
| _`repr_value`_ | Displays the value it receives in the _`build`_ method. Useful for debugging. |
<Admonition type="info" label="Tip">
Check out the [FlowRunner](../examples/flow-runner) example to understand how to call a flow from a custom component.
</Admonition>

8
docs/docs/components/embeddings.mdx
View file

@ -1,5 +1,13 @@
import Admonition from '@theme/Admonition';
# Embeddings
<Admonition type="caution" icon="🚧" title="ZONE UNDER CONSTRUCTION">
<p>
We appreciate your understanding as we polish our documentation – it may contain some rough edges. Share your feedback or report issues to help us improve! 🛠️📝
</p>
</Admonition>
Embeddings are vector representations of text that capture the semantic meaning of the text. They are created using text embedding models and allow us to think about the text in a vector space, enabling us to perform tasks like semantic search, where we look for pieces of text that are most similar in the vector space.
---

198
docs/docs/components/llms.mdx
View file

@ -1,2 +1,198 @@
import Admonition from '@theme/Admonition';
# LLMs
(coming soon)
<Admonition type="caution" icon="🚧" title="ZONE UNDER CONSTRUCTION">
<p>
We appreciate your understanding as we polish our documentation – it may contain some rough edges. Share your feedback or report issues to help us improve! 🛠️📝
</p>
</Admonition>
An LLM stands for Large Language Model. It is a core component of Langflow and provides a standard interface for interacting with different LLMs from various providers such as OpenAI, Cohere, and HuggingFace. LLMs are used widely throughout Langflow, including in chains and agents. They can be used to generate text based on a given prompt (or input).
---
### Anthropic
Wrapper around Anthropic's large language models. Find out more at [Anthropic](https://www.anthropic.com).
- **anthropic_api_key:** Used to authenticate and authorize access to the Anthropic API.
- **anthropic_api_url:** Specifies the URL of the Anthropic API to connect to.
- **temperature:** Tunes the degree of randomness in text generations. Should be a non-negative value.
---
### ChatAnthropic
Wrapper around Anthropic's large language model used for chat-based interactions. Find out more at [Anthropic](https://www.anthropic.com).
- **anthropic_api_key:** Used to authenticate and authorize access to the Anthropic API.
- **anthropic_api_url:** Specifies the URL of the Anthropic API to connect to.
- **temperature:** Tunes the degree of randomness in text generations. Should be a non-negative value.
---
### CTransformers
The `CTransformers` component provides access to the Transformer models implemented in C/C++ using the [GGML](https://github.com/ggerganov/ggml) library.
:::info
Make sure to have the `ctransformers` python package installed. Learn more about installation, supported models, and usage [here](https://github.com/marella/ctransformers).
:::
**config:** Configuration for the Transformer models. Check out [config](https://github.com/marella/ctransformers#config). Defaults to:
```
{
"top_k": 40,
"top_p": 0.95,
"temperature": 0.8,
"repetition_penalty": 1.1,
"last_n_tokens": 64,
"seed": -1,
"max_new_tokens": 256,
"stop": null,
"stream": false,
"reset": true,
"batch_size": 8,
"threads": -1,
"context_length": -1,
"gpu_layers": 0
}
```
**model:** The path to a model file or directory or the name of a Hugging Face Hub model repo.
**model_file:** The name of the model file in the repo or directory.
**model_type:** Transformer model to be used. Learn more [here](https://github.com/marella/ctransformers).
---
### ChatOpenAI
Wrapper around [OpenAI's](https://openai.com) chat large language models. This component supports some of the LLMs (Large Language Models) available by OpenAI and is used for tasks such as chatbots, Generative Question-Answering (GQA), and summarization.
- **max_tokens:** The maximum number of tokens to generate in the completion. `-1` returns as many tokens as possible, given the prompt and the model's maximal context size – defaults to `256`.
- **model_kwargs:** Holds any model parameters valid for creating non-specified calls.
- **model_name:** Defines the OpenAI chat model to be used.
- **openai_api_base:** Used to specify the base URL for the OpenAI API. It is typically set to the API endpoint provided by the OpenAI service.
- **openai_api_key:** Key used to authenticate and access the OpenAI API.
- **temperature:** Tunes the degree of randomness in text generations. Should be a non-negative value – defaults to `0.7`.
---
### Cohere
Wrapper around [Cohere's](https://cohere.com) large language models.
- **cohere_api_key:** Holds the API key required to authenticate with the Cohere service.
- **max_tokens:** Maximum number of tokens to predict per generation – defaults to `256`.
- **temperature:** Tunes the degree of randomness in text generations. Should be a non-negative value – defaults to `0.75`.
---
### HuggingFaceHub
Wrapper around [HuggingFace](https://www.huggingface.co/models) models.
:::info
The HuggingFace Hub is an online platform that hosts over 120k models, 20k datasets, and 50k demo apps, all of which are open-source and publicly available. Discover more at [HuggingFace](http://www.huggingface.co).
:::
- **huggingfacehub_api_token:** Token needed to authenticate the API.
- **model_kwargs:** Keyword arguments to pass to the model.
- **repo_id:** Model name to use – defaults to `gpt2`.
- **task:** Task to call the model with. Should be a task that returns `generated_text` or `summary_text`.
---
### LlamaCpp
The `LlamaCpp` component provides access to the `llama.cpp` models.
:::info
Make sure to have the `llama.cpp` python package installed. Learn more about installation, supported models, and usage [here](https://github.com/ggerganov/llama.cpp).
:::
- **echo:** Whether to echo the prompt – defaults to `False`.
- **f16_kv:** Use half-precision for key/value cache – defaults to `True`.
- **last_n_tokens_size:** The number of tokens to look back at when applying the repeat_penalty. Defaults to `64`.
- **logits_all:** Return logits for all tokens, not just the last token Defaults to `False`.
- **logprobs:** The number of logprobs to return. If None, no logprobs are returned.
- **lora_base:** The path to the Llama LoRA base model.
- **lora_path:** The path to the Llama LoRA. If None, no LoRa is loaded.
- **max_tokens:** The maximum number of tokens to generate. Defaults to `256`.
- **model_path:** The path to the Llama model file.
- **n_batch:** Number of tokens to process in parallel. Should be a number between 1 and n_ctx. Defaults to `8`.
- **n_ctx:** Token context window. Defaults to `512`.
- **n_gpu_layers:** Number of layers to be loaded into GPU memory. Default None.
- **n_parts:**Number of parts to split the model into. If -1, the number of parts is automatically determined. Defaults to `-1`.
- **n_threads:** Number of threads to use. If None, the number of threads is automatically determined.
- **repeat_penalty:** The penalty to apply to repeated tokens. Defaults to `1.1`.
- **seed:** Seed. If -1, a random seed is used. Defaults to `-1`.
- **stop:** A list of strings to stop generation when encountered.
- **streaming:** Whether to stream the results, token by token. Defaults to `True`.
- **suffix:** A suffix to append to the generated text. If None, no suffix is appended.
- **tags:** Tags to add to the run trace.
- **temperature:** The temperature to use for sampling. Defaults to `0.8`.
- **top_k:** The top-k value to use for sampling. Defaults to `40`.
- **top_p:** The top-p value to use for sampling. Defaults to `0.95`.
- **use_mlock:** Force the system to keep the model in RAM. Defaults to `False`.
- **use_mmap:** Whether to keep the model loaded in RAM. Defaults to `True`.
- **verbose:** This parameter is used to control the level of detail in the output of the chain. When set to True, it will print out some internal states of the chain while it is being run, which can help debug and understand the chain's behavior. If set to False, it will suppress the verbose output. Defaults to `False`.
- **vocab_only:** Only load the vocabulary, no weights. Defaults to `False`.
---
### OpenAI
Wrapper around [OpenAI's](https://openai.com) large language models.
- **max_tokens:** The maximum number of tokens to generate in the completion. `-1` returns as many tokens as possible, given the prompt and the model's maximal context size – defaults to `256`.
- **model_kwargs:** Holds any model parameters valid for creating non-specified calls.
- **model_name:** Defines the OpenAI model to be used.
- **openai_api_base:** Used to specify the base URL for the OpenAI API. It is typically set to the API endpoint provided by the OpenAI service.
- **openai_api_key:** Key used to authenticate and access the OpenAI API.
- **temperature:** Tunes the degree of randomness in text generations. Should be a non-negative value – defaults to `0.7`.
---
### VertexAI
Wrapper around [Google Vertex AI](https://cloud.google.com/vertex-ai) large language models.
:::info
Vertex AI is a cloud computing platform offered by Google Cloud Platform (GCP). It provides access, management, and development of applications and services through global data centers. To use Vertex AI PaLM, you need to have the [google-cloud-aiplatform](https://pypi.org/project/google-cloud-aiplatform/) Python package installed and credentials configured for your environment.
:::
- **credentials:** The default custom credentials (google.auth.credentials.Credentials) to use.
- **location:** The default location to use when making API calls – defaults to `us-central1`.
- **max_output_tokens:** Token limit determines the maximum amount of text output from one prompt – defaults to `128`.
- **model_name:** The name of the Vertex AI large language model – defaults to `text-bison`.
- **project:** The default GCP project to use when making Vertex API calls.
- **request_parallelism:** The amount of parallelism allowed for requests issued to VertexAI models – defaults to `5`.
- **temperature:** Tunes the degree of randomness in text generations. Should be a non-negative value – defaults to `0`.
- **top_k:** How the model selects tokens for output, the next token is selected from – defaults to `40`.
- **top_p:** Tokens are selected from most probable to least until the sum of their – defaults to `0.95`.
- **tuned_model_name:** The name of a tuned model. If provided, model_name is ignored.
- **verbose:** This parameter is used to control the level of detail in the output of the chain. When set to True, it will print out some internal states of the chain while it is being run, which can help debug and understand the chain's behavior. If set to False, it will suppress the verbose output – defaults to `False`.

10
docs/docs/components/loaders.mdx
View file

@ -1,2 +1,10 @@
import Admonition from '@theme/Admonition';
# Loaders
(coming soon)
<Admonition type="caution" icon="🚧" title="ZONE UNDER CONSTRUCTION">
<p>
We appreciate your understanding as we polish our documentation – it may contain some rough edges. Share your feedback or report issues to help us improve! 🛠️📝
</p>
</Admonition>

108
docs/docs/components/memories.mdx
View file

@ -1,2 +1,108 @@
import Admonition from '@theme/Admonition';
# Memories
(coming soon)
<Admonition type="caution" icon="🚧" title="ZONE UNDER CONSTRUCTION">
<p>
We appreciate your understanding as we polish our documentation – it may contain some rough edges. Share your feedback or report issues to help us improve! 🛠️📝
</p>
</Admonition>
Memory is a concept in chat-based applications that allows the system to remember previous interactions. It helps in maintaining the context of the conversation and enables the system to understand new messages in relation to past messages.
---
### ConversationBufferMemory
The `ConversationBufferMemory` component is a type of memory system that plainly stores the last few inputs and outputs of a conversation.
**Params**
- **input_key:** Used to specify the key under which the user input will be stored in the conversation memory. It allows you to provide the user's input to the chain for processing and generating a response.
- **memory_key:** Specifies the prompt variable name where the memory will store and retrieve the chat messages. It allows for the preservation of the conversation history throughout the interaction with the language model – defaults to `chat_history`.
- **output_key:** Used to specify the key under which the generated response will be stored in the conversation memory. It allows you to retrieve the response using the specified key.
- **return_messages:** Determines whether the history should be returned as a string or as a list of messages. If `return_messages` is set to True, the history will be returned as a list of messages. If `return_messages` is set to False or not specified, the history will be returned as a string. The default is `False`.
---
### ConversationBufferWindowMemory
`ConversationBufferWindowMemory` is a variation of the `ConversationBufferMemory` that maintains a list of the recent interactions in a conversation. It only keeps the last K interactions in memory, which can be useful for maintaining a sliding window of the most recent interactions without letting the buffer get too large.
**Params**
- **input_key:** Used to specify the keys in the memory object where the input messages should be stored. It allows for the retrieval and manipulation of input messages.
- **memory_key:** Specifies the prompt variable name where the memory will store and retrieve the chat messages. It allows for the preservation of the conversation history throughout the interaction with the language model. Defaults to `chat_history`.
- **k:** Used to specify the number of interactions or messages that should be stored in the conversation buffer. It determines the size of the sliding window that keeps track of the most recent interactions.
- **output_key:** Used to specify the key under which the generated response will be stored in the conversation memory. It allows you to retrieve the response using the specified key.
- **return_messages:** Determines whether the history should be returned as a string or as a list of messages. If `return_messages` is set to True, the history will be returned as a list of messages. If `return_messages` is set to False or not specified, the history will be returned as a string. The default is `False`.
---
### ConversationEntityMemory
The `ConversationEntityMemory` component incorporates intricate memory structures, specifically a key-value store, for entities referenced in a conversation. This facilitates the storage and retrieval of information related to entities that have been mentioned throughout the conversation.
**Params**
- **Entity Store:** Structure that stores information about specific entities mentioned in a conversation.
- **LLM:** Language Model to use in the `ConversationEntityMemory`.
- **chat_history_key:** Specify a unique identifier for the chat history data associated with a particular entity. This allows for organizing and accessing the chat history data for each entity within the conversation entity memory. Defaults to `history`
- **input_key:** Used to specify the keys in the memory object where the input messages should be stored. It allows for the retrieval and manipulation of input messages.
- **k:** Refers to the number of entities that can be stored in the memory. It determines the maximum number of entities that can be stored and retrieved from the memory object. Defaults to `10`
- **output_key:** Used to specify the key under which the generated response will be stored in the conversation memory. It allows you to retrieve the response using the specified key.
- **return_messages:** Determines whether the history should be returned as a string or as a list of messages. If `return_messages` is set to True, the history will be returned as a list of messages. If `return_messages` is set to False or not specified, the history will be returned as a string. The default is `False`.
---
### ConversationKGMemory
`ConversationKGMemory` is a type of memory that uses a knowledge graph to recreate memory. It allows the extraction of entities and knowledge triplets from a new message, using previous messages as context.
**Params**
- **LLM:** Language Model to use in the `ConversationKGMemory`.
- **input_key:** Used to specify the keys in the memory object where the input messages should be stored. It allows for the retrieval and manipulation of input messages.
- **k:** Represents the number of previous conversation turns that will be stored in the memory. By setting "k" to 2, it means that the memory will retain the previous 2 conversation turns, allowing the model to access and utilize the information from those turns during the conversation. Defaults to `10`
- **memory_key:** Specifies the prompt variable name where the memory will store and retrieve the chat messages. It allows for the preservation of the conversation history throughout the interaction with the language model. Defaults to `chat_history`.
- **output_key:** Used to specify the key under which the generated response will be stored in the conversation memory. It allows you to retrieve the response using the specified key.
- **return_messages:** Determines whether the history should be returned as a string or as a list of messages. If `return_messages` is set to True, the history will be returned as a list of messages. If `return_messages` is set to False or not specified, the history will be returned as a string. The default is `False`.
---
### ConversationSummaryMemory
The `ConversationSummaryMemory` is a memory component that creates a summary of the conversation over time. It condenses information from the conversation and stores the current summary in memory. It is particularly useful for longer conversations where keeping the entire message history in the prompt would take up too many tokens.
**Params**
- **LLM:** Language Model to use in the `ConversationSummaryMemory`.
- **input_key:** Used to specify the keys in the memory object where the input messages should be stored. It allows for the retrieval and manipulation of input messages.
- **memory_key:** Specifies the prompt variable name where the memory will store and retrieve the chat messages. It allows for the preservation of the conversation history throughout the interaction with the language model. Defaults to `chat_history`.
- **output_key:** Used to specify the key under which the generated response will be stored in the conversation memory. It allows you to retrieve the response using the specified key.
- **return_messages:** Determines whether the history should be returned as a string or as a list of messages. If `return_messages` is set to True, the history will be returned as a list of messages. If `return_messages` is set to False or not specified, the history will be returned as a string. The default is `False`.
---
### PostgresChatMessageHistory
The `PostgresChatMessageHistory` is a memory component that allows for the storage and retrieval of chat message history using a PostgreSQL database. The connection to the PostgreSQL database is established using a connection string, which includes the necessary authentication and database information.
**Params**
- **connection_string:** Refers to a string that contains the necessary information to establish a connection to a PostgreSQL database. The `connection_string` typically includes details such as the username, password, host, port, and database name required to connect to the PostgreSQL database. Defaults to `postgresql://postgres:mypassword@localhost/chat_history`
- **session_id:** It is a unique identifier that is used to associate chat message history with a specific session or conversation.
- **table_name:** Refers to the name of the table in the PostgreSQL database where the chat message history will be stored. Defaults to `message_store`
---
### VectorRetrieverMemory
The `VectorRetrieverMemory` is a memory component that allows for the retrieval of vectors based on a given query. It is used to perform vector-based searches and retrievals.
**Params**
- **Retriever:** The retriever used to fetch documents.
- **input_key:** Used to specify the keys in the memory object where the input messages should be stored. It allows for the retrieval and manipulation of input messages.
- **memory_key:** Specifies the prompt variable name where the memory will store and retrieve the chat messages. It allows for the preservation of the conversation history throughout the interaction with the language model – defaults to `chat_history`.
- **return_messages:** Determines whether the history should be returned as a string or as a list of messages. If `return_messages` is set to True, the history will be returned as a list of messages. If `return_messages` is set to False or not specified, the history will be returned as a string – defaults to `False`.

20
docs/docs/components/prompts.mdx
View file

@ -1,5 +1,15 @@
import Admonition from "@theme/Admonition";
# Prompts
<Admonition type="caution" icon="🚧" title="ZONE UNDER CONSTRUCTION">
<p>
We appreciate your understanding as we polish our documentation – it may
contain some rough edges. Share your feedback or report issues to help us
improve! 🛠️📝
</p>
</Admonition>
A prompt refers to the input given to a language model. It is constructed from multiple components and can be parametrized using prompt templates. A prompt template is a reproducible way to generate prompts and allow for easy customization through input variables.
---
@ -8,8 +18,10 @@ A prompt refers to the input given to a language model. It is constructed from m
The `PromptTemplate` component allows users to create prompts and define variables that provide control over instructing the model. The template can take in a set of variables from the end user and generates the prompt once the conversation is initiated.
:::info
Once a variable is defined in the prompt template, it becomes a component input of its own. Check out [Prompt Customization](../guidelines/prompt-customization.mdx) to learn more.
:::
<Admonition type="info">
Once a variable is defined in the prompt template, it becomes a component
input of its own. Check out [Prompt
Customization](../guidelines/prompt-customization.mdx) to learn more.
</Admonition>
- **template:** Template used to format an individual request.
- **template:** Template used to format an individual request.

24
docs/docs/components/retrievers.mdx Normal file
View file

@ -0,0 +1,24 @@
import Admonition from '@theme/Admonition';
# Retrievers
<Admonition type="caution" icon="🚧" title="ZONE UNDER CONSTRUCTION">
<p>
We appreciate your understanding as we polish our documentation – it may contain some rough edges. Share your feedback or report issues to help us improve! 🛠️📝
</p>
</Admonition>
A retriever is an interface that returns documents given an unstructured query. It is more general than a vector store and does not need to be able to store documents, only to return or retrieve them.
---
### MultiQueryRetriever
The `MultiQueryRetriever` component automates the process of generating multiple queries, retrieves relevant documents for each query, and combines the results to provide a more extensive and diverse set of potentially relevant documents. This approach enhances the effectiveness of the retrieval process and helps overcome the limitations of traditional distance-based retrieval methods.
**Params**
- **LLM:** Language Model to use in the `MultiQueryRetriever`.
- **Prompt:** Prompt to represent a schema for an LLM.
- **Retriever:** The retriever used to fetch documents.
- **parser_key:** This parameter is used to specify the key or attribute name of the parsed output that will be used for retrieval. It determines how the results from the language model are split into a list of queries. Defaults to `lines`, which means that the output from the language model will be split into a list of lines of text. This allows the retriever to retrieve relevant documents based on each line of text separately.

49
docs/docs/components/text-splitters.mdx
View file

@ -1,2 +1,49 @@
import Admonition from '@theme/Admonition';
# Text Splitters
(coming soon)
<Admonition type="caution" icon="🚧" title="ZONE UNDER CONSTRUCTION">
<p>
We appreciate your understanding as we polish our documentation – it may contain some rough edges. Share your feedback or report issues to help us improve! 🛠️📝
</p>
</Admonition>
A text splitter is a tool that divides a document or text into smaller chunks or segments. It is used to break down large texts into more manageable pieces for analysis or processing.
---
### CharacterTextSplitter
The `CharacterTextSplitter` is used to split a long text into smaller chunks based on a specified character. It splits the text by trying to keep paragraphs, sentences, and words together as long as possible, as these are semantically related pieces of text.
**Params**
- **Documents:** Input documents to split.
- **chunk_overlap:** Determines the number of characters that overlap between consecutive chunks when splitting text. It specifies how much of the previous chunk should be included in the next chunk.
For example, if the `chunk_overlap` is set to 20 and the `chunk_size` is set to 100, the splitter will create chunks of 100 characters each, but the last 20 characters of each chunk will overlap with the first 20 characters of the next chunk. This allows for a smoother transition between chunks and ensures that no information is lost – defaults to `200`.
- **chunk_size:** Determines the maximum number of characters in each chunk when splitting a text. It specifies the size or length of each chunk.
For example, if the chunk_size is set to 100, the splitter will create chunks of 100 characters each. If the text is longer than 100 characters, it will be divided into multiple chunks of equal size, except for the last chunk, which may be smaller if there are remaining characters –defaults to `1000`.
- **separator:** Specifies the character that will be used to split the text into chunks – defaults to `.`
---
### RecursiveCharacterTextSplitter
The `RecursiveCharacterTextSplitter` splits the text by trying to keep paragraphs, sentences, and words together as long as possible, similar to the `CharacterTextSplitter`. However, it also recursively splits the text into smaller chunks if the chunk size exceeds a specified threshold.
**Params**
- **Documents:** Input documents to split.
- **chunk_overlap:** Determines the number of characters that overlap between consecutive chunks when splitting text. It specifies how much of the previous chunk should be included in the next chunk.
- **chunk_size:** Determines the maximum number of characters in each chunk when splitting a text. It specifies the size or length of each chunk.
- **separator_type:** The parameter allows the user to split the code with multiple language support. It supports various languages such as Text, Ruby, Python, Solidity, Java, and more. Defaults to `Text`.
- **separators:** The `separators` in RecursiveCharacterTextSplitter are the characters used to split the text into chunks. The text splitter tries to create chunks based on splitting on the first character in the list of `separators`. If any chunks are too large, it moves on to the next character in the list and continues splitting. Defaults to `.`

9
docs/docs/components/toolkits.mdx
View file

@ -1,2 +1,9 @@
import Admonition from '@theme/Admonition';
# Toolkits
(coming soon)
<Admonition type="caution" icon="🚧" title="ZONE UNDER CONSTRUCTION">
<p>
We appreciate your understanding as we polish our documentation – it may contain some rough edges. Share your feedback or report issues to help us improve! 🛠️📝
</p>
</Admonition>

9
docs/docs/components/tools.mdx
View file

@ -1,2 +1,9 @@
import Admonition from '@theme/Admonition';
# Tools
(coming soon)
<Admonition type="caution" icon="🚧" title="ZONE UNDER CONSTRUCTION">
<p>
We appreciate your understanding as we polish our documentation – it may contain some rough edges. Share your feedback or report issues to help us improve! 🛠️📝
</p>
</Admonition>

10
docs/docs/components/utilities.mdx
View file

@ -1,2 +1,10 @@
import Admonition from '@theme/Admonition';
# Utilities
(coming soon)
<Admonition type="caution" icon="🚧" title="ZONE UNDER CONSTRUCTION">
<p>
We appreciate your understanding as we polish our documentation – it may contain some rough edges. Share your feedback or report issues to help us improve! 🛠️📝
</p>
</Admonition>

9
docs/docs/components/vector-stores.mdx
View file

@ -1,2 +1,9 @@
import Admonition from '@theme/Admonition';
# Vector Stores
(coming soon)
<Admonition type="caution" icon="🚧" title="ZONE UNDER CONSTRUCTION">
<p>
We appreciate your understanding as we polish our documentation – it may contain some rough edges. Share your feedback or report issues to help us improve! 🛠️📝
</p>
</Admonition>

20
docs/docs/components/wrappers.mdx
View file

@ -1,2 +1,20 @@
import Admonition from '@theme/Admonition';
# Wrappers
(coming soon)
<Admonition type="caution" icon="🚧" title="ZONE UNDER CONSTRUCTION">
<p>
We appreciate your understanding as we polish our documentation – it may contain some rough edges. Share your feedback or report issues to help us improve! 🛠️📝
</p>
</Admonition>
### TextRequestsWrapper
This component is designed to work with the Python Requests module, which is a popular tool for making web requests. Used to fetch data from a particular website.
**Params**
- **header:** specifies the headers to be included in the HTTP request. Defaults to `{'Authorization': 'Bearer <token>'}`.
Headers are key-value pairs that provide additional information about the request or the client making the request. They can be used to send authentication credentials, specify the content type of the request, set cookies, and more. They allow the client and the server to communicate additional information beyond the basic request.

21
docs/docs/contributing/how-contribute.md
View file

@ -36,10 +36,9 @@ Before you start, make sure you have the following installed:
- Poetry (>=1.4)
- Node.js
Then install the dependencies and start the development server for the backend:
Then, in the root folder, install the dependencies and start the development server for the backend:
```bash
make install_backend
make backend
```
@ -49,6 +48,7 @@ And the frontend:
make frontend
```
---
## Docker compose
@ -59,4 +59,19 @@ The following snippet will run the backend and frontend in separate containers.
docker compose up --build
# or
make dev build=1
```
```
---
## Documentation
The documentation is built using [Docusaurus](https://docusaurus.io/). To run the documentation locally, run the following commands:
```bash
cd docs
npm install
npm run start
```
The documentation will be available at `localhost:3000` and all the files are located in the `docs/docs` folder.
Once you are done with your changes, you can create a Pull Request to the `main` branch.

7
docs/docs/examples/buffer-memory.mdx
View file

@ -1,3 +1,5 @@
import Admonition from "@theme/Admonition";
# Buffer Memory
For certain applications, retaining past interactions is crucial. For that, chains and agents may accept a memory component as one of their input parameters. The `ConversationBufferMemory` component is one of them. It stores messages and extracts them into variables.
@ -17,9 +19,10 @@ import ZoomableImage from "/src/theme/ZoomableImage.js";
#### <a target="\_blank" href="json_files/Buffer_Memory.json" download>Download Flow</a>
:::note LangChain Components 🦜🔗
<Admonition type="note" title="LangChain Components 🦜🔗">
- [`ConversationBufferMemory`](https://python.langchain.com/docs/modules/memory/how_to/buffer)
- [`ConversationChain`](https://python.langchain.com/docs/modules/chains/)
- [`ChatOpenAI`](https://python.langchain.com/docs/modules/model_io/models/chat/integrations/openai)
:::
</Admonition>

13
docs/docs/examples/conversation-chain.mdx
View file

@ -1,10 +1,14 @@
import Admonition from "@theme/Admonition";
# Conversation Chain
This example shows how to instantiate a simple `ConversationChain` component using a Language Model (LLM). Once the Node Status turns green 🟢, the chat will be ready to take in user messages. Here, we used `ChatOpenAI` to act as the required LLM input, but you can use any LLM for this purpose.
:::info
<Admonition type="info">
Make sure to always get the API key from the provider.
:::
</Admonition>
## ⛓️ Langflow Example
@ -21,8 +25,9 @@ import ZoomableImage from "/src/theme/ZoomableImage.js";
#### <a target="\_blank" href="json_files/Basic_Chat.json" download>Download Flow</a>
:::note LangChain Components 🦜🔗
<Admonition type="note" title="LangChain Components 🦜🔗">
- [`ConversationChain`](https://python.langchain.com/docs/modules/chains/)
- [`ChatOpenAI`](https://python.langchain.com/docs/modules/model_io/models/chat/integrations/openai)
:::
</Admonition>

24
docs/docs/examples/csv-loader.mdx
View file

@ -1,3 +1,5 @@
import Admonition from "@theme/Admonition";
# CSV Loader
The `VectoStoreAgent` component retrieves information from one or more vector stores. This example shows a `VectoStoreAgent` connected to a CSV file through the `Chroma` vector store. Process description:
@ -7,13 +9,18 @@ The `VectoStoreAgent` component retrieves information from one or more vector st
- These chunks feed the `Chroma` vector store, which converts them into vectors and stores them for fast indexing.
- Finally, the agent accesses the information of the vector store through the `VectorStoreInfo` tool.
:::info
The vector store is used for efficient semantic search, while `VectorStoreInfo` carries information about it, such as its name and description. Embeddings are a way to represent words, phrases, or any entities in a vector space. Learn more about them [here](https://platform.openai.com/docs/guides/embeddings/what-are-embeddings).
:::
<Admonition type="info">
The vector store is used for efficient semantic search, while
`VectorStoreInfo` carries information about it, such as its name and
description. Embeddings are a way to represent words, phrases, or any entities
in a vector space. Learn more about them
[here](https://platform.openai.com/docs/guides/embeddings/what-are-embeddings).
</Admonition>
:::tip
Once you build this flow, ask questions about the data in the chat interface (e.g., number of rows or columns).
:::
<Admonition type="tip">
Once you build this flow, ask questions about the data in the chat interface
(e.g., number of rows or columns).
</Admonition>
## ⛓️ Langflow Example
@ -30,7 +37,7 @@ import ZoomableImage from "/src/theme/ZoomableImage.js";
#### <a target="\_blank" href="json_files/CSV_Loader.json" download>Download Flow</a>
:::note LangChain Components 🦜🔗
<Admonition type="note" title="LangChain Components 🦜🔗">
- [`CSVLoader`](https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/csv)
- [`CharacterTextSplitter`](https://python.langchain.com/docs/modules/data_connection/document_transformers/text_splitters/character_text_splitter)
@ -39,4 +46,5 @@ import ZoomableImage from "/src/theme/ZoomableImage.js";
- [`VectorStoreInfo`](https://python.langchain.com/docs/modules/data_connection/vectorstores/)
- [`OpenAI`](https://python.langchain.com/docs/modules/model_io/models/llms/integrations/openai)
- [`VectorStoreAgent`](https://python.langchain.com/docs/modules/agents/toolkits/vectorstore)
:::
</Admonition>

365
docs/docs/examples/flow-runner.mdx Normal file
View file

@ -0,0 +1,365 @@
---
description: Custom Components
hide_table_of_contents: true
---
import ZoomableImage from "/src/theme/ZoomableImage.js";
import Admonition from "@theme/Admonition";
# FlowRunner Component
The CustomComponent class allows us to create components that interact with Langflow itself. In this example, we will make a component that runs other flows available in "My Collection".
<ZoomableImage
alt="Document Processor Component"
sources={{
light: "img/flow_runner.png",
}}
style={{
width: "30%",
margin: "0 auto",
display: "flex",
justifyContent: "center",
}}
/>
We will cover how to:
- List Collection flows using the _`list_flows`_ method.
- Load a flow using the _`load_flow`_ method.
- Configure a dropdown input field using the _`options`_ parameter.
<details open>
<summary>Example Code</summary>
```python
from langflow import CustomComponent
from langchain.schema import Document
class FlowRunner(CustomComponent):
display_name = "Flow Runner"
description = "Run other flows using a document as input."
def build_config(self):
flows = self.list_flows()
flow_names = [f.name for f in flows]
return {"flow_name": {"options": flow_names,
"display_name": "Flow Name",
},
"document": {"display_name": "Document"}
}
def build(self, flow_name: str, document: Document) -> Document:
# List the flows
flows = self.list_flows()
# Get the flow that matches the selected name
# You can also get the flow by id
# using self.get_flow(flow_id=flow_id)
tweaks = {}
flow = self.get_flow(flow_name=flow_name, tweaks=tweaks)
# Get the page_content from the document
if document and isinstance(document, list):
document = document[0]
page_content = document.page_content
# Use it in the flow
result = flow(page_content)
return Document(page_content=str(result))
```
</details>
<CH.Scrollycoding rows={20} className={""}>
```python
from langflow import CustomComponent
class MyComponent(CustomComponent):
display_name = "Custom Component"
description = "This is a custom component"
def build_config(self):
...
def build(self):
...
```
The typical structure of a Custom Component is composed of _`display_name`_ and _`description`_ attributes, _`build`_ and _`build_config`_ methods.
---
```python
from langflow import CustomComponent
# focus
class FlowRunner(CustomComponent):
# focus
display_name = "Flow Runner"
# focus
description = "Run other flows"
def build_config(self):
...
def build(self):
...
```
Let's start by defining our component's _`display_name`_ and _`description`_.
---
```python
from langflow import CustomComponent
# focus
from langchain.schema import Document
class FlowRunner(CustomComponent):
display_name = "Flow Runner"
description = "Run other flows using a document as input."
def build_config(self):
...
def build(self):
...
```
Second, we will import _`Document`_ from the [_langchain.schema_](https://docs.langchain.com/docs/components/schema/) module. This will be the return type of the _`build`_ method.
---
```python
from langflow import CustomComponent
# focus
from langchain.schema import Document
class FlowRunner(CustomComponent):
display_name = "Flow Runner"
description = "Run other flows using a document as input."
def build_config(self):
...
# focus
def build(self, flow_name: str, document: Document) -> Document:
...
```
Now, let's add the [parameters](focus://11[20:55]) and the [return type](focus://11[60:69]) to the _`build`_ method. The parameters added are:
- _`flow_name`_ is the name of the flow we want to run.
- _`document`_ is the input document to be passed to that flow.
- Since _`Document`_ is a Langchain type, it will add an input [handle](../guidelines/components) to the component ([see more](../components/custom)).
---
```python focus=13:14
from langflow import CustomComponent
from langchain.schema import Document
class FlowRunner(CustomComponent):
display_name = "Flow Runner"
description = "Run other flows using a document as input."
def build_config(self):
...
def build(self, flow_name: str, document: Document) -> Document:
# List the flows
flows = self.list_flows()
```
We can now start writing the _`build`_ method. Let's list available flows in "My Collection" using the _`list_flows`_ method.
---
```python focus=15:18
from langflow import CustomComponent
from langchain.schema import Document
class FlowRunner(CustomComponent):
display_name = "Flow Runner"
description = "Run other flows using a document as input."
def build_config(self):
...
def build(self, flow_name: str, document: Document) -> Document:
# List the flows
flows = self.list_flows()
# Get the flow that matches the selected name
# You can also get the flow by id
# using self.get_flow(flow_id=flow_id)
tweaks = {}
flow = self.get_flow(flow_name=flow_name, tweaks=tweaks)
```
And retrieve a flow that matches the selected name (we'll make a dropdown input field for the user to choose among flow names).
<Admonition type="caution">
From version 0.4.0, names are unique, which was not the case in previous
versions. This might lead to unexpected results if using flows with the same
name.
</Admonition>
---
```python
from langflow import CustomComponent
from langchain.schema import Document
class FlowRunner(CustomComponent):
display_name = "Flow Runner"
description = "Run other flows using a document as input."
def build_config(self):
...
def build(self, flow_name: str, document: Document) -> Document:
# List the flows
flows = self.list_flows()
# Get the flow that matches the selected name
# You can also get the flow by id
# using self.get_flow(flow_id=flow_id)
tweaks = {}
flow = self.get_flow(flow_name=flow_name, tweaks=tweaks)
```
You can load this flow using _`get_flow`_ and set a _`tweaks`_ dictionary to customize it. Find more about tweaks in our [features guidelines](../guidelines/features#code).
---
```python
from langflow import CustomComponent
from langchain.schema import Document
class FlowRunner(CustomComponent):
display_name = "Flow Runner"
description = "Run other flows using a document as input."
def build_config(self):
...
def build(self, flow_name: str, document: Document) -> Document:
# List the flows
flows = self.list_flows()
# Get the flow that matches the selected name
# You can also get the flow by id
# using self.get_flow(flow_id=flow_id)
tweaks = {}
flow = self.get_flow(flow_name=flow_name, tweaks=tweaks)
# Get the page_content from the document
if document and isinstance(document, list):
document = document[0]
page_content = document.page_content
# Use it in the flow
result = flow(page_content)
return Document(page_content=str(result))
```
We are using a _`Document`_ as input because it is a straightforward way to pass text data in Langflow (specifically because you can connect it to many [loaders](../components/loaders)).
Generally, a flow will take a string or a dictionary as input because that's what LangChain components expect.
In case you are passing a dictionary, you need to build it according to the needs of the flow you are using.
The content of a document can be extracted using the _`page_content`_ attribute, which is a string, and passed as an argument to the selected flow.
---
```python focus=9:16
from langflow import CustomComponent
from langchain.schema import Document
class FlowRunner(CustomComponent):
display_name = "Flow Runner"
description = "Run other flows using a document as input."
def build_config(self):
flows = self.list_flows()
flow_names = [f.name for f in flows]
return {"flow_name": {"options": flow_names,
"display_name": "Flow Name",
},
"document": {"display_name": "Document"}
}
def build(self, flow_name: str, document: Document) -> Document:
# List the flows
flows = self.list_flows()
# Get the flow that matches the selected name
# You can also get the flow by id
# using self.get_flow(flow_id=flow_id)
tweaks = {}
flow = self.get_flow(flow_name=flow_name, tweaks=tweaks)
# Get the page_content from the document
if document and isinstance(document, list):
document = document[0]
page_content = document.page_content
# Use it in the flow
result = flow(page_content)
return Document(page_content=str(result))
```
Finally, we can add field customizations through the _`build_config`_ method. Here we added the _`options`_ key to make the _`flow_name`_ field a dropdown menu. Check out the [custom component reference](../components/custom) for a list of available keys.
<Admonition type="caution">
Make sure that the field type is _`str`_ and _`options`_ values are strings.
</Admonition>
</CH.Scrollycoding>
Done! This is what our script and custom component looks like:
<div style={{
display: "flex",
justifyContent: "center",
}}>
<ZoomableImage
alt="Document Processor Code"
sources={{
light: "img/flow_runner_code.png",
}}
style={{
maxWidth: "100%",
margin: "0 auto",
display: "flex",
justifyContent: "center",
}}
/>
<ZoomableImage
alt="Document Processor Component"
sources={{
light: "img/flow_runner.png",
}}
style={{
width: "40%",
margin: "0 auto",
display: "flex",
justifyContent: "center",
}}
/>
</div>

18
docs/docs/examples/how-upload-examples.mdx
View file

@ -7,16 +7,14 @@ import ZoomableImage from "/src/theme/ZoomableImage.js";
We welcome all examples that can help our community learn and explore Langflow's capabilities.
Langflow Examples is a repository on [GitHub](https://github.com/logspace-ai/langflow_examples) that contains examples of flows that people can use for inspiration and learning.
<div
style={{ marginBottom: "20px", display: "flex", justifyContent: "center" }}
>
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: "img/community-examples.png",
}}
/>
</div>
{" "}
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: "img/community-examples.png",
}}
style={{ width: "100%" }}
/>
To upload examples, please follow these steps:

15
docs/docs/examples/midjourney-prompt-chain.mdx
View file

@ -1,3 +1,5 @@
import Admonition from "@theme/Admonition";
# MidJourney Prompt Chain
The `MidJourneyPromptChain` can be used to generate imaginative and detailed MidJourney prompts.
@ -14,9 +16,11 @@ And get a response such as:
Imagine a mysterious forest, the trees are tall and ancient, their branches reaching up to the sky. Through the darkness, a dragon emerges from the shadows, its scales shimmering in the moonlight. Its wingspan is immense, and its eyes glow with a fierce intensity. It is a majestic and powerful creature, one that commands both respect and fear.
```
:::tip
Notice that the `ConversationSummaryMemory` stores a summary of the conversation over time. Try using it to create better prompts as the conversation goes on.
:::
<Admonition type="tip">
Notice that the `ConversationSummaryMemory` stores a summary of the
conversation over time. Try using it to create better prompts as the
conversation goes on.
</Admonition>
## ⛓️ Langflow Example
@ -33,8 +37,9 @@ import ZoomableImage from "/src/theme/ZoomableImage.js";
#### <a target="\_blank" href="json_files/MidJourney_Prompt_Chain.json" download>Download Flow</a>
:::note LangChain Components 🦜🔗
<Admonition type="note" title="LangChain Components 🦜🔗">
- [`OpenAI`](https://python.langchain.com/docs/modules/model_io/models/llms/integrations/openai)
- [`ConversationSummaryMemory`](https://python.langchain.com/docs/modules/memory/how_to/summary)
:::
</Admonition>

29
docs/docs/examples/multiple-vectorstores.mdx
View file

@ -1,26 +1,31 @@
import Admonition from "@theme/Admonition";
# Multiple Vector Stores
The example below shows an agent operating with two vector stores built upon different data sources.
The `TextLoader` loads a TXT file, while the `WebBaseLoader` pulls text from webpages into a document format to accessed downstream. The `Chroma` vector stores are created analogous to what we have demonstrated in our [CSV Loader](/examples/csv-loader.mdx) example. Finally, the `VectorStoreRouterAgent` constructs an agent that routes between the vector stores.
:::info
Get the TXT file used [here](https://github.com/hwchase17/chat-your-data/blob/master/state_of_the_union.txt).
:::
<Admonition type="info">
Get the TXT file used
[here](https://github.com/hwchase17/chat-your-data/blob/master/state_of_the_union.txt).
</Admonition>
URL used by the `WebBaseLoader`:
```txt
```text
https://pt.wikipedia.org/wiki/Harry_Potter
```
:::tip
When you build the flow, request information about one of the sources. The agent should be able to use the correct source to generate a response.
:::
<Admonition type="tip">
When you build the flow, request information about one of the sources. The
agent should be able to use the correct source to generate a response.
</Admonition>
:::info
Learn more about Multiple Vector Stores [here](https://python.langchain.com/docs/modules/agents/toolkits/vectorstore?highlight=Multiple%20Vector%20Stores#multiple-vectorstores).
:::
<Admonition type="info">
Learn more about Multiple Vector Stores
[here](https://python.langchain.com/docs/modules/agents/toolkits/vectorstore?highlight=Multiple%20Vector%20Stores#multiple-vectorstores).
</Admonition>
## ⛓️ Langflow Example
@ -37,7 +42,7 @@ import ZoomableImage from "/src/theme/ZoomableImage.js";
#### <a target="\_blank" href="json_files/Multiple_Vector_Stores.json" download>Download Flow</a>
:::note LangChain Components 🦜🔗
<Admonition type="note" title="LangChain Components 🦜🔗">
- [`WebBaseLoader`](https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/web_base)
- [`TextLoader`](https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/unstructured_file)
@ -49,4 +54,4 @@ import ZoomableImage from "/src/theme/ZoomableImage.js";
- [`VectorStoreRouterToolkit`](https://python.langchain.com/docs/modules/agents/toolkits/vectorstore)
- [`VectorStoreRouterAgent`](https://python.langchain.com/docs/modules/agents/toolkits/vectorstore)
:::
</Admonition>

23
docs/docs/examples/python-function.mdx
View file

@ -1,3 +1,5 @@
import Admonition from "@theme/Admonition";
# Python Function
Langflow allows you to create a customized tool using the `PythonFunction` connected to a `Tool` component. In this example, Regex is used in Python to validate a pattern.
@ -15,15 +17,19 @@ def is_brazilian_zipcode(zipcode: str) -> bool:
return False
```
:::tip
When a tool is called, it is often desirable to have its output returned directly to the user. You can do this by setting the **return_direct** flag for a tool to be True.
:::
<Admonition type="tip">
When a tool is called, it is often desirable to have its output returned
directly to the user. You can do this by setting the **return_direct** flag
for a tool to be True.
</Admonition>
The `AgentInitializer` component is a quick way to construct an agent from the model and tools.
:::info
The `PythonFunction` is a custom component that uses the LangChain 🦜🔗 tool decorator. Learn more about it [here](https://python.langchain.com/docs/modules/agents/tools/how_to/custom_tools).
:::
<Admonition type="info">
The `PythonFunction` is a custom component that uses the LangChain 🦜🔗 tool
decorator. Learn more about it
[here](https://python.langchain.com/docs/modules/agents/tools/how_to/custom_tools).
</Admonition>
## ⛓️ Langflow Example
@ -40,9 +46,10 @@ import ZoomableImage from "/src/theme/ZoomableImage.js";
#### <a target="\_blank" href="json_files/Python_Function.json" download>Download Flow</a>
:::note LangChain Components 🦜🔗
<Admonition type="note" title="LangChain Components 🦜🔗">
- [`PythonFunctionTool`](https://python.langchain.com/docs/modules/agents/tools/how_to/custom_tools)
- [`ChatOpenAI`](https://python.langchain.com/docs/modules/model_io/models/chat/integrations/openai)
- [`AgentInitializer`](https://python.langchain.com/docs/modules/agents/)
:::
</Admonition>

28
docs/docs/examples/serp-api-tool.mdx
View file

@ -1,24 +1,29 @@
import Admonition from "@theme/Admonition";
# Serp API Tool
The [Serp API](https://serpapi.com/) (Search Engine Results Page) allows developers to scrape results from search engines such as Google, Bing and Yahoo, and can be used as in Langflow through the `Search` component.
:::info
To use the Serp API, you first need to sign up [Serp API](https://serpapi.com/) for an API key on the provider's website.
:::
<Admonition type="info">
To use the Serp API, you first need to sign up [Serp
API](https://serpapi.com/) for an API key on the provider's website.
</Admonition>
Here, the `ZeroShotPrompt` component specifies a prompt template for the `ZeroShotAgent`. Set a _Prefix_ and _Suffix_ with rules for the agent to obey. In the example, we used default templates.
The `LLMChain` is a simple chain that takes in a prompt template, formats it with the user input, and returns the response from an LLM.
:::tip
In this example, we used [`ChatOpenAI`](https://platform.openai.com/) as the LLM, but feel free to experiment with other Language Models!
:::
<Admonition type="tip">
In this example, we used [`ChatOpenAI`](https://platform.openai.com/) as the
LLM, but feel free to experiment with other Language Models!
</Admonition>
The `ZeroShotAgent` takes the `LLMChain` and the `Search` tool as inputs, using the tool to find information when necessary.
:::info
Learn more about the Serp API [here](https://python.langchain.com/docs/modules/agents/tools/integrations/serpapi).
:::
<Admonition type="info">
Learn more about the Serp API
[here](https://python.langchain.com/docs/modules/agents/tools/integrations/serpapi).
</Admonition>
## ⛓️ Langflow Example
@ -35,11 +40,12 @@ import ZoomableImage from "/src/theme/ZoomableImage.js";
#### <a target="\_blank" href="json_files/SerpAPI_Tool.json" download>Download Flow</a>
:::note LangChain Components 🦜🔗
<Admonition type="note" title="LangChain Components 🦜🔗">
- [`ZeroShotPrompt`](https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/)
- [`OpenAI`](https://python.langchain.com/docs/modules/model_io/models/llms/integrations/openai)
- [`LLMChain`](https://python.langchain.com/docs/modules/chains/foundational/llm_chain)
- [`Search`](https://python.langchain.com/docs/modules/agents/tools/integrations/serpapi)
- [`ZeroShotAgent`](https://python.langchain.com/docs/modules/agents/how_to/custom_mrkl_agent)
:::
</Admonition>

19
docs/docs/getting-started/hugging-face-spaces.mdx
View file

@ -6,15 +6,14 @@ import ThemedImage from "@theme/ThemedImage";
import useBaseUrl from "@docusaurus/useBaseUrl";
import ZoomableImage from "/src/theme/ZoomableImage.js";
<div
style={{ marginBottom: "20px", display: "flex", justifyContent: "center" }}
>
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: "img/hugging-face.png",
}}
/>
</div>
{" "}
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: "img/hugging-face.png",
}}
style={{ width: "100%" }}
/>
Check out Langflow on [HuggingFace Spaces](https://huggingface.co/spaces/Logspace/Langflow).

78
docs/docs/guidelines/chat-interface.mdx
View file

@ -7,58 +7,46 @@ import ReactPlayer from "react-player";
Langflow’s chat interface provides a user-friendly experience and functionality to interact with the model and customize the prompt. The sidebar brings options that allow users to view and edit pre-defined prompt variables. This feature facilitates quick experimentation by enabling the modification of variable values right in the chat.
<div
style={{ marginBottom: "20px", display: "flex", justifyContent: "center" }}
>
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: useBaseUrl("img/chat_interface.png"),
}}
style={{ width: "100%", maxWidth: "800px", margin: "0 auto" }}
/>
</div>
{" "}
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: useBaseUrl("img/chat_interface.png"),
}}
style={{ width: "100%", maxWidth: "800px", margin: "0 auto" }}
/>
Notice that editing variables in the chat interface take place temporarily and won’t change their original value in the components once the chat is closed.
<div
style={{ marginBottom: "20px", display: "flex", justifyContent: "center" }}
>
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: useBaseUrl("img/chat_interface2.png"),
}}
style={{ width: "100%", maxWidth: "800px", margin: "0 auto" }}
/>
</div>
{" "}
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: useBaseUrl("img/chat_interface2.png"),
}}
style={{ width: "100%", maxWidth: "800px", margin: "0 auto" }}
/>
To view the complete prompt in its original, structured format, click the "Display Prompt" option. This feature lets you see the prompt exactly as it entered the model.
<div
style={{ marginBottom: "20px", display: "flex", justifyContent: "center" }}
>
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: useBaseUrl("img/chat_interface3.png"),
}}
style={{ width: "100%", maxWidth: "800px", margin: "0 auto" }}
/>
</div>
{" "}
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: useBaseUrl("img/chat_interface3.png"),
}}
style={{ width: "100%", maxWidth: "800px", margin: "0 auto" }}
/>
In the chat interface, you can redefine which variable should be interpreted as the chat input. This gives you control over these inputs and allows dynamic and creative interactions.
<div
style={{ marginBottom: "20px", display: "flex", justifyContent: "center" }}
>
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: useBaseUrl("img/chat_interface4.png"),
}}
style={{ width: "100%", maxWidth: "800px", margin: "0 auto" }}
/>
</div>
{" "}
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: useBaseUrl("img/chat_interface4.png"),
}}
style={{ width: "100%", maxWidth: "800px", margin: "0 auto" }}
/>

209
docs/docs/guidelines/chat-widget.mdx Normal file
View file

@ -0,0 +1,209 @@
import ThemedImage from "@theme/ThemedImage";
import useBaseUrl from "@docusaurus/useBaseUrl";
import ZoomableImage from "/src/theme/ZoomableImage.js";
import ReactPlayer from "react-player";
import Admonition from "@theme/Admonition";
# Chat Widget
<div style={{ marginBottom: "20px" }}>
The <b>Langflow Chat Widget</b> is a powerful web component that enables
communication with a Langflow project. This widget allows for a chat interface
embedding, allowing the integration of Langflow into web applications
effortlessly.
</div>
## Features
🌟 **Seamless Integration:** Easily integrate the Langflow Chat Widget into your website or web application with just a few lines of JavaScript.
🚀 **Interactive Chat Interface:** Engage your users with a user-friendly conversation, powered by Langflow's advanced language understanding capabilities.
🎛️ **Customizable Styling:** Customize the appearance of the chat widget to match your application's design and branding.
🌐 **Multilingual Support:** Communicate with users in multiple languages, opening up your application to a global audience.
---
## Usage
<div style={{ marginBottom: "20px" }}>
You can get the HTML code embedded with the chat by clicking the Code button
at the Sidebar after building a flow.
</div>
{" "}
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: useBaseUrl("img/widget-sidebar.png"),
}}
style={{ width: "50%", maxWidth: "600px", margin: "0 auto" }}
/>
<div style={{ marginBottom: "20px" }}>
Clicking the Chat Widget HTML tab, you'll get the code to be inserted. Read
below to learn how to use it with HTML, React and Angular.
</div>
{" "}
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: useBaseUrl("img/widget-code.png"),
}}
style={{ width: "100%", maxWidth: "800px", margin: "0 auto" }}
/>
---
### HTML
The Chat Widget can be embedded into any HTML page, inside a _`<body>`_ tag, as demonstrated in the video below.
<div
style={{ marginBottom: "20px", display: "flex", justifyContent: "center" }}
>
<ReactPlayer playing controls url="/videos/langflow_widget.mp4" />
</div>
---
### React
To embed the Chat Widget using React, you'll need to insert this _`<script>`_ tag into the React _index.html_ file, inside the _`<body>`_ tag:
```html
<script src="https://cdn.jsdelivr.net/gh/logspace-ai/langflow-embedded-chat@main/dist/build/static/js/bundle.min.js"></script>
```
Then, declare your Web Component and encapsulate it in a React component.
```jsx
declare global {
namespace JSX {
interface IntrinsicElements {
"langflow-chat": any;
}
}
}
export default function ChatWidget({ className }) {
return (
<div className={className}>
<langflow-chat
chat_inputs='{"your_key":"value"}'
chat_input_field="your_chat_key"
flow_id="your_flow_id"
host_url="langflow_url"
></langflow-chat>
</div>
);
}
```
Finally, you can place the component anywhere in your code to display the Chat Widget.
---
### Angular
To use it in Angular, first add this _`<script>`_ tag into the Angular _index.html_ file, inside the _`<body>`_ tag.
```html
<script src="https://cdn.jsdelivr.net/gh/logspace-ai/langflow-embedded-chat@main/dist/build/static/js/bundle.min.js"></script>
```
When you use a custom web component in an Angular template, the Angular compiler might show a warning when it doesn't recognize the custom elements by default. To suppress this warning, add _`CUSTOM_ELEMENTS_SCHEMA`_ to the module's _`@NgModule.schemas`_.
- Open the module file (it typically ends with _.module.ts_) where you'd add the _`langflow-chat`_ web component.
- Import _`CUSTOM_ELEMENTS_SCHEMA`_ at the top of the file:
```ts
import { NgModule, CUSTOM_ELEMENTS_SCHEMA } from "@angular/core";
```
- Add _`CUSTOM_ELEMENTS_SCHEMA`_ to the 'schemas' array inside the '@NgModule' decorator:
```ts
@NgModule({
declarations: [
// ... Other components and directives ...
],
imports: [
// ... Other imported modules ...
],
schemas: [CUSTOM_ELEMENTS_SCHEMA], // Add the CUSTOM_ELEMENTS_SCHEMA here
})
export class YourModule {}
```
In your Angular project, find the component belonging to the module where _`CUSTOM_ELEMENTS_SCHEMA`_ was added.
- Inside the template, add the _`langflow-chat`_ tag to include the Chat Widget in your component's view:
```jsx
<langflow-chat
chat_inputs='{"your_key":"value"}'
chat_input_field="your_chat_key"
flow_id="your_flow_id"
host_url="langflow_url"
></langflow-chat>
```
<Admonition type="info">
<ul>
<li>
_`CUSTOM_ELEMENTS_SCHEMA`_ is a built-in schema that allows Angular to
recognize custom elements.
</li>
<li>
Adding _`CUSTOM_ELEMENTS_SCHEMA`_ tells Angular to allow custom elements
in your templates, and it will suppress the warning related to unknown
elements like _`langflow-chat`_.
</li>
<li>
Notice that you can only use the Chat Widget in components that are part
of the module where you added _`CUSTOM_ELEMENTS_SCHEMA`_.
</li>
</ul>
</Admonition>
---
## Configuration
Use the widget API to customize your Chat Widget:
<Admonition type="caution">
Props with the type JSON need to be passed as Stringified JSONs, with the
format &#123;<span>"key":"value"</span>&#125;.
</Admonition>
| Prop | Type | Required | Description |
| --------------------- | ------- | -------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| bot_message_style | JSON | No | Applies custom formatting to bot messages. |
| chat_input_field | String | Yes | Defines the type of the input field for chat messages. |
| chat_inputs | JSON | Yes | Determines the chat input elements and their respective values. |
| chat_output_key | String | No | Specifies which output to display if multiple outputs are available. |
| chat_position | String | No | Positions the chat window on the screen (options include: top-left, top-center, top-right, center-left, center-right, bottom-right, bottom-center, bottom-left). |
| chat_trigger_style | JSON | No | Styles the chat trigger button. |
| chat_window_style | JSON | No | Customizes the overall appearance of the chat window. |
| error_message_style | JSON | No | Sets the format for error messages within the chat window. |
| flow_id | String | Yes | Identifies the flow that the component is associated with. |
| height | Number | No | Sets the height of the chat window in pixels. |
| host_url | String | Yes | Specifies the URL of the host for chat component communication. |
| input_container_style | JSON | No | Applies styling to the container where chat messages are entered. |
| input_style | JSON | No | Sets the style for the chat input field. |
| online | Boolean | No | Toggles the online status of the chat component. |
| online_message | String | No | Sets a custom message to display when the chat component is online. |
| placeholder | String | No | Sets the placeholder text for the chat input field. |
| placeholder_sending | String | No | Sets the placeholder text to display while a message is being sent. |
| send_button_style | JSON | No | Sets the style for the send button in the chat window. |
| send_icon_style | JSON | No | Sets the style for the send icon in the chat window. |
| tweaks | JSON | No | Applies additional custom adjustments for the associated flow. |
| user_message_style | JSON | No | Determines the formatting for user messages in the chat window. |
| width | Number | No | Sets the width of the chat window in pixels. |
| window_title | String | No | Sets the title displayed in the chat window's header or title bar. |

19
docs/docs/guidelines/components.mdx
View file

@ -25,17 +25,14 @@ Components are the building blocks of the flows. They are made of inputs, output
of that type is required.
</div>
<div
style={{ marginBottom: "20px", display: "flex", justifyContent: "center" }}
>
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: useBaseUrl("img/single-compenent.png"),
}}
style={{ width: "100%", maxWidth: "800px", margin: "0 auto" }}
/>
</div>
{" "}
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: useBaseUrl("img/single-compenent.png"),
}}
style={{ width: "100%", maxWidth: "800px", margin: "0 auto" }}
/>
<div style={{ marginBottom: "20px" }}>
On the top right corner, you will find the component status icon 🔴. Make the

407
docs/docs/guidelines/custom-component.mdx Normal file
View file

@ -0,0 +1,407 @@
---
description: Custom Components
hide_table_of_contents: true
---
import ZoomableImage from "/src/theme/ZoomableImage.js";
import Admonition from "@theme/Admonition";
# Custom Components
In Langflow, a Custom Component is a special component type that allows users to extend the platform's functionality by creating their own reusable and configurable components.
A Custom Component is created from a user-defined Python script that uses the _`CustomComponent`_ class provided by the Langflow library. These components can be as simple as a basic function that takes and returns a string or as complex as a combination of multiple sub-components and API calls.
Let's take a look at the basic rules and features. Then we'll go over an example.
## TL;DR
- Create a class that inherits from _`CustomComponent`_ and contains a _`build`_ method.
- Use arguments with [Type Annotations (or Type Hints)](https://docs.python.org/3/library/typing.html) of the _`build`_ method to create component fields.
- Use the _`build_config`_ method to customize how these fields look and behave.
- Set up a folder with your components to load them up in Langflow's sidebar.
Here is an example:
<div style={{
display: "flex",
justifyContent: "center",
}}>
<CH.Code linuNumbers={false}>
```python
from langflow import CustomComponent
from langchain.schema import Document
class DocumentProcessor(CustomComponent):
display_name = "Document Processor"
description = "This component processes a document"
def build_config(self) -> dict:
options = ["Uppercase", "Lowercase", "Titlecase"]
return {
"function": {"options": options,
"value": options[0]}}
def build(self, document: Document, function: str) -> Document:
if isinstance(document, list):
document = document[0]
page_content = document.page_content
if function == "Uppercase":
page_content = page_content.upper()
elif function == "Lowercase":
page_content = page_content.lower()
elif function == "Titlecase":
page_content = page_content.title()
self.repr_value = f"Result of {function} function: {page_content}"
return Document(page_content=page_content)
```
</CH.Code>
<ZoomableImage
alt="Document Processor Component"
sources={{
light: "img/document_processor.png",
}}
style={{
width: "40%",
margin: "0 auto",
display: "flex",
justifyContent: "center",
}}
/>
</div>
<Admonition type="tip">
Check out [FlowRunner Component](../examples/flow-runner) for a more complex
example.
</Admonition>
---
## Rules
The Python script for every Custom Component should follow a set of rules. Let's go over them one by one:
<CH.Scrollycoding rows={20} className={""}>
### Rule 1
The script must contain a **single class** that inherits from _`CustomComponent`_.
```python
from langflow import CustomComponent
from langchain.schema import Document
class MyComponent(CustomComponent):
display_name = "Custom Component"
description = "This is a custom component"
def build_config(self) -> dict:
...
def build(self, document: Document, function: str) -> Document:
...
```
---
### Rule 2
This class requires a _`build`_ method used to run the component and define its fields.
```python
from langflow import CustomComponent
from langchain.schema import Document
class MyComponent(CustomComponent):
display_name = "Custom Component"
description = "This is a custom component"
def build_config(self) -> dict:
...
# focus
# mark
def build(self) -> Document:
...
```
---
The [Return Type Annotation](https://docs.python.org/3/library/typing.html) of the _`build`_ method defines the component type (e.g., Chain, BaseLLM, or basic Python types). Check out all supported types in the [component reference](../components/custom).
```python
from langflow import CustomComponent
from langchain.schema import Document
class MyComponent(CustomComponent):
display_name = "Custom Component"
description = "This is a custom component"
def build_config(self) -> dict:
...
# focus[20:31]
# mark
def build(self) -> Document:
...
```
---
```python
from langflow import CustomComponent
from langchain.schema import Document
class MyComponent(CustomComponent):
display_name = "Custom Component"
description = "This is a custom component"
def build_config(self) -> dict:
...
def build(self) -> Document:
...
```
### Rule 3
The class can have a [_`build_config`_](focus://8) method, which defines configuration fields for the component. The [_`build_config`_](focus://8) method should always return a dictionary with specific keys representing the field names and their corresponding configurations. It must follow the format described below:
- Top-level keys are field names.
- Their values are also of type _`dict`_. They specify the behavior of the generated fields.
Check out the [component reference](../components/custom) for more details on the available field configurations.
---
```python
from langflow import CustomComponent
from langchain.schema import Document
class MyComponent(CustomComponent):
display_name = "Custom Component"
description = "This is a custom component"
def build_config(self) -> dict:
...
def build(self) -> Document:
...
```
## Example
Let's create a custom component that processes a document (_`langchain.schema.Document`_) using a simple function.
---
### Pick a display name
To start, let's choose a name for our component by adding a _`display_name`_ attribute. This name will appear on the canvas. The name of the class is not relevant, but let's call it _`DocumentProcessor`_.
```python
from langflow import CustomComponent
from langchain.schema import Document
# focus
class DocumentProcessor(CustomComponent):
# focus
display_name = "Document Processor"
description = "This is a custom component"
def build_config(self) -> dict:
...
def build(self) -> Document:
...
```
---
### Write a description
We can also write a description for it using a _`description`_ attribute.
```python
from langflow import CustomComponent
from langchain.schema import Document
class DocumentProcessor(CustomComponent):
display_name = "Document Processor"
description = "This component processes a document"
def build_config(self) -> dict:
...
def build(self) -> Document:
...
```
---
```python
from langflow import CustomComponent
from langchain.schema import Document
class DocumentProcessor(CustomComponent):
display_name = "Document Processor"
description = "This component processes a document"
def build_config(self) -> dict:
...
def build(self, document: Document, function: str) -> Document:
if isinstance(document, list):
document = document[0]
page_content = document.page_content
if function == "Uppercase":
page_content = page_content.upper()
elif function == "Lowercase":
page_content = page_content.lower()
elif function == "Titlecase":
page_content = page_content.title()
self.repr_value = f"Result of {function} function: {page_content}"
return Document(page_content=page_content)
```
### Add the build method
Here, the build method takes two input parameters: _`document`_, representing the input document to be processed, and _`function`_, a string representing the selected text transformation to be applied (either "Uppercase," "Lowercase," or "Titlecase"). The method processes the text content of the input Document based on the selected function.
The attribute _`repr_value`_ is used to display the result of the component on the canvas. It is optional and can be used to display any string value.
The return type is _`Document`_.
---
### Customize the component fields
The _`build_config`_ method is here defined to customize the component fields.
- _`options`_ determines that the field will be a dropdown menu. The list values and field type must be _`str`_.
- _`value`_ is the default option of the dropdown menu.
- _`display_name`_ is the name of the field to be displayed.
```python
from langflow import CustomComponent
from langchain.schema import Document
class DocumentProcessor(CustomComponent):
display_name = "Document Processor"
description = "This component processes a document"
def build_config(self) -> dict:
options = ["Uppercase", "Lowercase", "Titlecase"]
return {
"function": {"options": options,
"value": options[0],
"display_name": "Function"
},
"document": {"display_name": "Document"}
}
def build(self, document: Document, function: str) -> Document:
if isinstance(document, list):
document = document[0]
page_content = document.page_content
if function == "Uppercase":
page_content = page_content.upper()
elif function == "Lowercase":
page_content = page_content.lower()
elif function == "Titlecase":
page_content = page_content.title()
self.repr_value = f"Result of {function} function: {page_content}"
return Document(page_content=page_content)
```
</CH.Scrollycoding>
All done! This is what our script and brand-new custom component look like:
<div style={{
display: "flex",
justifyContent: "center",
}}>
<ZoomableImage
alt="Document Processor Code"
sources={{
light: "img/document_processor_code.png",
}}
style={{
maxWidth: "100%",
margin: "0 auto",
display: "flex",
justifyContent: "center",
}}
/>
<ZoomableImage
alt="Document Processor Component"
sources={{
light: "img/document_processor.png",
}}
style={{
width: "40%",
margin: "0 auto",
display: "flex",
justifyContent: "center",
}}
/>
</div>
---
## Loading Custom Components
For advanced customization, Langflow offers the option to create and load custom components outside of the standard interface. This process involves creating the desired components using a text editor and loading them using the Langflow CLI.
### Folder Structure
Create a folder that follows the same structural conventions as the [config.yaml](https://github.com/logspace-ai/langflow/blob/dev/src/backend/langflow/config.yaml) file. Inside this main directory, use a `custom_components` subdirectory for your custom components.
Inside `custom_components`, you can create a Python file for each component. Similarly, any custom agents should be housed in an `agents` subdirectory.
If you use a subdirectory name that is not in our config.yaml file, your component will appear in an `Other` category in the sidebar.
Your structure should look something like this:
```
.
└── custom_components
├── document_processor.py
└── ...
└── agents
└── ...
└── my_agents <-- Other category
└── ...
```
### Loading Custom Components
You can specify the path to your custom components using the _`--components-path`_ argument when running the Langflow CLI, as shown below:
```bash
langflow --components-path /path/to/components
```
Alternatively, you can set the `LANGFLOW_COMPONENTS_PATH` environment variable:
```bash
export LANGFLOW_COMPONENTS_PATH=/path/to/components
langflow
```
Langflow will attempt to load all of the components found in the specified directory. If a component fails to load due to errors in the component's code, Langflow will print an error message to the console but will continue loading the rest of the components.
### Interacting with Custom Components
Once your custom components have been loaded successfully, they will appear in Langflow's sidebar. From there, you can add them to your Langflow canvas for use. However, please note that components with errors will not be available for addition to the canvas. Always ensure your code is error-free before attempting to load components.
Remember, creating custom components allows you to extend the functionality of Langflow to better suit your unique needs. Happy coding!

25
docs/docs/guidelines/features.mdx
View file

@ -2,6 +2,7 @@ import ThemedImage from "@theme/ThemedImage";
import useBaseUrl from "@docusaurus/useBaseUrl";
import ZoomableImage from "/src/theme/ZoomableImage.js";
import ReactPlayer from "react-player";
import Admonition from "@theme/Admonition";
# Features
@ -12,17 +13,14 @@ import ReactPlayer from "react-player";
below:
</div>
<div
style={{ marginBottom: "20px", display: "flex", justifyContent: "center" }}
>
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: useBaseUrl("img/features.png"),
}}
style={{ width: "100%", maxWidth: "800px", margin: "0 auto" }}
/>
</div>
{" "}
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: useBaseUrl("img/features.png"),
}}
style={{ width: "100%", maxWidth: "800px", margin: "0 auto" }}
/>
<div style={{ marginBottom: "20px" }}>
Further down, we will explain each of these options.
@ -34,9 +32,10 @@ import ReactPlayer from "react-player";
Flows can be exported and imported as JSON files.
:::caution
<Admonition type="caution">
Watch out for API keys being stored in local files.
:::
</Admonition>
---

100
docs/docs/guidelines/prompt-customization.mdx
View file

@ -7,80 +7,62 @@ import ReactPlayer from "react-player";
The prompt template allows users to create prompts and define variables that provide control over instructing the model.
<div
style={{ marginBottom: "20px", display: "flex", justifyContent: "center" }}
>
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: useBaseUrl("img/prompt_customization.png"),
}}
style={{ width: "100%", maxWidth: "800px", margin: "0 auto" }}
/>
</div>
{" "}
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: useBaseUrl("img/prompt_customization.png"),
}}
style={{ width: "100%", maxWidth: "800px", margin: "0 auto" }}
/>
Variables can be used to define instructions, questions, context, inputs, or examples for the model and can be created with any chosen name in curly brackets, e.g., `{variable_name}`. They act as placeholders for parts of the text that can be easily modified.
<div
style={{ marginBottom: "20px", display: "flex", justifyContent: "center" }}
>
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: useBaseUrl("img/prompt_customization2.png"),
}}
style={{ width: "100%", maxWidth: "800px", margin: "0 auto" }}
/>
</div>
{" "}
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: useBaseUrl("img/prompt_customization2.png"),
}}
style={{ width: "100%", maxWidth: "800px", margin: "0 auto" }}
/>
Once inserted, these variables are immediately recognized as new fields in the prompt component. Here, you can define their values within the component itself or leave a field empty to be adjusted over the chat interface.
<div
style={{ marginBottom: "20px", display: "flex", justifyContent: "center" }}
>
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: useBaseUrl("img/prompt_customization3.png"),
}}
style={{ width: "100%", maxWidth: "800px", margin: "0 auto" }}
/>
</div>
{" "}
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: useBaseUrl("img/prompt_customization3.png"),
}}
style={{ width: "100%", maxWidth: "800px", margin: "0 auto" }}
/>
You can also use documents or output parsers as prompt variables. By plugging them into prompt handles, they’ll disable and feed that input field.
<div
style={{ marginBottom: "20px", display: "flex", justifyContent: "center" }}
>
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: useBaseUrl("img/prompt_customization4.png"),
}}
style={{ width: "100%", maxWidth: "800px", margin: "0 auto" }}
/>
</div>
{" "}
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: useBaseUrl("img/prompt_customization4.png"),
}}
style={{ width: "100%", maxWidth: "800px", margin: "0 auto" }}
/>
With this, users can interact with documents, webpages, or any other type of content directly from the prompt, which allows for seamless integration of external resources with the language model.
If working with an interactive (chat-like) flow, remember to keep one of the input variables empty to behave as the chat input.
<div
style={{ marginBottom: "20px", display: "flex", justifyContent: "center" }}
>
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: useBaseUrl("img/prompt_customization5.png"),
}}
style={{ width: "100%", maxWidth: "800px", margin: "0 auto" }}
/>
</div>
{" "}
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: useBaseUrl("img/prompt_customization5.png"),
}}
style={{ width: "100%", maxWidth: "800px", margin: "0 auto" }}
/>

3
docs/docs/guides/chatprompttemplate_guide.mdx
View file

@ -39,8 +39,7 @@ In this guide, we will modify the "Basic Chat with Prompt and History" example,
5. Open the "Prompt" field on the SystemMessagePromptTemplate component.
6. Enter the text: `You are a {role} that {behavior}.`
6. Enter the text: _`You are a {role} that {behavior}.`_
7. Save your changes by clicking on "Check & Save".
8. Define the 'role' variable by typing "obedient assistant".

18
docs/docs/index.mdx
View file

@ -6,13 +6,11 @@ import ThemedImage from "@theme/ThemedImage";
import useBaseUrl from "@docusaurus/useBaseUrl";
import ZoomableImage from "/src/theme/ZoomableImage.js";
<div
style={{ marginBottom: "20px", display: "flex", justifyContent: "center" }}
>
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: "img/new_langflow2.gif",
}}
/>
</div>
{" "}
<ZoomableImage
alt="Docusaurus themed image"
sources={{
light: "img/new_langflow2.gif",
}}
style={{ width: "100%" }}
/>

256
docs/docusaurus.config.js
View file

@ -1,127 +1,145 @@
const lightCodeTheme = require("prism-react-renderer/themes/github");
const { remarkCodeHike } = require("@code-hike/mdx");
// With JSDoc @type annotations, IDEs can provide config autocompletion
/** @type {import('@docusaurus/types').DocusaurusConfig} */
(
module.exports = {
title: "Langflow Documentation",
tagline: "Langflow is a GUI for LangChain, designed with react-flow",
favicon: "img/favicon.ico",
url: "https://logspace-ai.github.io",
baseUrl: "/",
onBrokenLinks: "throw",
onBrokenMarkdownLinks: "warn",
organizationName: "logspace-ai",
projectName: "langflow",
trailingSlash: false,
customFields: {
mendableAnonKey: process.env.MENDABLE_ANON_KEY,
},
i18n: {
defaultLocale: "en",
locales: ["en"],
},
presets: [
[
"@docusaurus/preset-classic",
/** @type {import('@docusaurus/preset-classic').Options} */
({
docs: {
routeBasePath: "/",
sidebarPath: require.resolve("./sidebars.js"),
path: "docs",
// sidebarPath: 'sidebars.js',
},
theme: {
customCss: require.resolve("./src/css/custom.css"),
},
}),
],
],
plugins: [
["docusaurus-node-polyfills", { excludeAliases: ["console"] }],
"docusaurus-plugin-image-zoom",
// ....
async function myPlugin(context, options) {
return {
name: "docusaurus-tailwindcss",
configurePostCss(postcssOptions) {
// Appends TailwindCSS and AutoPrefixer.
postcssOptions.plugins.push(require("tailwindcss"));
postcssOptions.plugins.push(require("autoprefixer"));
return postcssOptions;
},
};
},
],
themeConfig:
/** @type {import('@docusaurus/preset-classic').ThemeConfig} */
module.exports = {
title: "Langflow Documentation",
tagline: "Langflow is a GUI for LangChain, designed with react-flow",
favicon: "img/favicon.ico",
url: "https://logspace-ai.github.io",
baseUrl: "/",
onBrokenLinks: "throw",
onBrokenMarkdownLinks: "warn",
organizationName: "logspace-ai",
projectName: "langflow",
trailingSlash: false,
customFields: {
mendableAnonKey: process.env.MENDABLE_ANON_KEY,
},
i18n: {
defaultLocale: "en",
locales: ["en"],
},
presets: [
[
"@docusaurus/preset-classic",
/** @type {import('@docusaurus/preset-classic').Options} */
({
navbar: {
hideOnScroll: true,
title: "Langflow",
logo: {
alt: "Langflow",
src: "img/chain.png",
},
items: [
// right
{
position: "right",
href: "https://github.com/logspace-ai/langflow",
position: "right",
className: "header-github-link",
target: "_blank",
rel: null,
},
{
position: "right",
href: "https://twitter.com/logspace_ai",
position: "right",
className: "header-twitter-link",
target: "_blank",
rel: null,
},
{
position: "right",
href: "https://discord.gg/EqksyE2EX9",
position: "right",
className: "header-discord-link",
target: "_blank",
rel: null,
},
docs: {
beforeDefaultRemarkPlugins: [
[
remarkCodeHike,
{
theme: "github-light",
showCopyButton: true,
lineNumbers: true,
},
],
],
routeBasePath: "/",
sidebarPath: require.resolve("./sidebars.js"),
path: "docs",
// sidebarPath: 'sidebars.js',
},
theme: {
customCss: [
require.resolve("@code-hike/mdx/styles.css"),
require.resolve("./src/css/custom.css"),
],
},
tableOfContents: {
minHeadingLevel: 2,
maxHeadingLevel: 5,
},
colorMode: {
defaultMode: "light",
disableSwitch: true,
respectPrefersColorScheme: false,
},
announcementBar: {
content:
'⭐️ If you like ⛓️Langflow, star it on <a target="_blank" rel="noopener noreferrer" href="https://github.com/logspace-ai/langflow">GitHub</a>! ⭐️',
backgroundColor: "#B53D38", //Mustard Yellow #D19900 #D4B20B - Salmon #E9967A
textColor: "#fff",
isCloseable: false,
},
footer: {
links: [],
copyright: `Copyright © ${new Date().getFullYear()} Logspace.`,
},
zoom: {
selector: ".markdown :not(a) > img:not(.no-zoom)",
background: {
light: "rgba(240, 240, 240, 0.9)",
},
config: {},
},
prism: {
theme: lightCodeTheme,
},
}),
}
);
],
],
plugins: [
["docusaurus-node-polyfills", { excludeAliases: ["console"] }],
"docusaurus-plugin-image-zoom",
// ....
async function myPlugin(context, options) {
return {
name: "docusaurus-tailwindcss",
configurePostCss(postcssOptions) {
// Appends TailwindCSS and AutoPrefixer.
postcssOptions.plugins.push(require("tailwindcss"));
postcssOptions.plugins.push(require("autoprefixer"));
return postcssOptions;
},
};
},
],
themes: ["mdx-v2"],
themeConfig:
/** @type {import('@docusaurus/preset-classic').ThemeConfig} */
({
navbar: {
hideOnScroll: true,
title: "Langflow",
logo: {
alt: "Langflow",
src: "img/chain.png",
},
items: [
// right
{
position: "right",
href: "https://github.com/logspace-ai/langflow",
position: "right",
className: "header-github-link",
target: "_blank",
rel: null,
},
{
position: "right",
href: "https://twitter.com/logspace_ai",
position: "right",
className: "header-twitter-link",
target: "_blank",
rel: null,
},
{
position: "right",
href: "https://discord.gg/EqksyE2EX9",
position: "right",
className: "header-discord-link",
target: "_blank",
rel: null,
},
],
},
tableOfContents: {
minHeadingLevel: 2,
maxHeadingLevel: 5,
},
colorMode: {
defaultMode: "light",
disableSwitch: true,
respectPrefersColorScheme: false,
},
announcementBar: {
content:
'⭐️ If you like ⛓️Langflow, star it on <a target="_blank" rel="noopener noreferrer" href="https://github.com/logspace-ai/langflow">GitHub</a>! ⭐️',
backgroundColor: "#E8EBF1", //Mustard Yellow #D19900 #D4B20B - Salmon #E9967A
textColor: "#1C1E21",
isCloseable: false,
},
footer: {
links: [],
copyright: `Copyright © ${new Date().getFullYear()} Logspace.`,
},
zoom: {
selector: ".markdown :not(a) > img:not(.no-zoom)",
background: {
light: "rgba(240, 240, 240, 0.9)",
},
config: {},
},
// prism: {
// theme: require("prism-react-renderer/themes/dracula"),
// },
docs: {
sidebar: {
hideable: true,
},
},
}),
};

2033
docs/package-lock.json generated
View file

File diff suppressed because it is too large Load diff

6
docs/package.json
View file

@ -15,12 +15,13 @@
},
"dependencies": {
"@babel/preset-react": "^7.22.3",
"@code-hike/mdx": "^0.9.0",
"@docusaurus/core": "2.4.1",
"@docusaurus/plugin-ideal-image": "^2.4.1",
"@docusaurus/preset-classic": "2.4.1",
"@docusaurus/theme-classic": "^2.4.1",
"@docusaurus/theme-search-algolia": "^2.4.1",
"@mdx-js/react": "^1.6.22",
"@mdx-js/react": "^2.3.0",
"@mendable/search": "^0.0.114",
"@pbe/react-yandex-maps": "^1.2.4",
"@prismicio/client": "^7.0.1",
@ -28,6 +29,7 @@
"autoprefixer": "^10.4.14",
"clsx": "^1.2.1",
"docusaurus-plugin-image-zoom": "^0.1.4",
"docusaurus-theme-mdx-v2": "^0.1.2",
"jquery": "^3.7.0",
"medium-zoom": "^1.0.8",
"node-fetch": "^3.3.1",
@ -67,4 +69,4 @@
"engines": {
"node": ">=16.14"
}
}
}

5
docs/sidebars.js
View file

@ -21,6 +21,8 @@ module.exports = {
"guidelines/collection",
"guidelines/prompt-customization",
"guidelines/chat-interface",
"guidelines/chat-widget",
"guidelines/custom-component",
],
},
{
@ -30,11 +32,13 @@ module.exports = {
items: [
"components/agents",
"components/chains",
"components/custom",
"components/embeddings",
"components/llms",
"components/loaders",
"components/memories",
"components/prompts",
"components/retrievers",
"components/text-splitters",
"components/toolkits",
"components/tools",
@ -63,6 +67,7 @@ module.exports = {
label: "Examples",
collapsed: false,
items: [
"examples/flow-runner",
"examples/conversation-chain",
"examples/buffer-memory",
"examples/midjourney-prompt-chain",

41
docs/src/css/custom.css
View file

@ -3,17 +3,19 @@
* bundles Infima by default. Infima is a CSS framework designed to
* work well for content-centric websites.
*/
:root {
:root {
--ifm-background-color: var(--token-primary-bg-c);
--ifm-navbar-link-hover-color: initial;
--ifm-navbar-padding-vertical: 0;
--ifm-navbar-item-padding-vertical: 0;
--ifm-font-family-base: -apple-system, BlinkMacSystemFont, Inter, Helvetica, Arial, sans-serif, 'Apple Color Emoji', 'Segoe UI emoji';
--ifm-font-family-monospace: 'SFMono-Regular', 'Roboto Mono', Consolas, 'Liberation Mono', Menlo, Courier, monospace;
--ifm-font-family-base: -apple-system, BlinkMacSystemFont, Inter, Helvetica,
Arial, sans-serif, "Apple Color Emoji", "Segoe UI emoji";
--ifm-font-family-monospace: "SFMono-Regular", "Roboto Mono", Consolas,
"Liberation Mono", Menlo, Courier, monospace;
}
.theme-doc-sidebar-item-category.menu__list-item:not(:first-child) {
margin-top: 1.5rem!important;
margin-top: 1.5rem !important;
}
.docusaurus-highlight-code-line {
@ -31,7 +33,7 @@
transform: skewY(6deg);
}
[class^='announcementBar'] {
[class^="announcementBar"] {
z-index: 10;
}
@ -112,7 +114,7 @@ body {
}
.header-github-link:before {
content: '';
content: "";
width: 24px;
height: 24px;
display: flex;
@ -126,7 +128,7 @@ body {
}
.header-twitter-link::before {
content: '';
content: "";
width: 24px;
height: 24px;
display: flex;
@ -140,7 +142,7 @@ body {
}
.header-discord-link::before {
content: '';
content: "";
width: 24px;
height: 24px;
display: flex;
@ -148,7 +150,6 @@ body {
background-size: contain;
}
/* Images */
.image-rendering-crisp {
image-rendering: crisp-edges;
@ -164,7 +165,7 @@ body {
.img-center {
display: flex;
justify-content: center;
width: 100%,
width: 100%;
}
.resized-image {
@ -188,4 +189,22 @@ body {
.mendable-search {
width: 140px;
}
}
}
/*
.ch-scrollycoding {
gap: 10rem !important;
} */
.ch-scrollycoding-content {
max-width: 55% !important;
min-width: 40% !important;
}
.ch-scrollycoding-sticker {
max-width: 60% !important;
min-width: 45% !important;
}
.ch-scrollycoding-step-content {
min-height: 70px;
}

28
docs/src/theme/ZoomableImage.js
View file

@ -1,8 +1,9 @@
import React, { useState, useEffect } from 'react';
import ThemedImage from '@theme/ThemedImage';
import useBaseUrl from '@docusaurus/useBaseUrl';
import React, { useState, useEffect } from "react";
import ThemedImage from "@theme/ThemedImage";
import useBaseUrl from "@docusaurus/useBaseUrl";
const ZoomableImage = ({ alt, sources }) => {
const ZoomableImage = ({ alt, sources, style }) => {
// add style here
const [isFullscreen, setIsFullscreen] = useState(false);
const toggleFullscreen = () => {
@ -10,27 +11,36 @@ const ZoomableImage = ({ alt, sources }) => {
};
const handleKeyPress = (event) => {
if (event.key === 'Escape') {
if (event.key === "Escape") {
setIsFullscreen(false);
}
};
useEffect(() => {
if (isFullscreen) {
document.addEventListener('keydown', handleKeyPress);
document.addEventListener("keydown", handleKeyPress);
} else {
document.removeEventListener('keydown', handleKeyPress);
document.removeEventListener("keydown", handleKeyPress);
}
return () => {
document.removeEventListener('keydown', handleKeyPress);
document.removeEventListener("keydown", handleKeyPress);
};
}, [isFullscreen]);
// Default style
const defaultStyle = {
width: "50%",
margin: "0 auto",
display: "flex",
justifyContent: "center",
};
return (
<div
className={`zoomable-image ${isFullscreen ? 'fullscreen' : ''}`}
className={`zoomable-image ${isFullscreen ? "fullscreen" : ""}`}
onClick={toggleFullscreen}
style={{ ...defaultStyle, ...style }}
>
<ThemedImage
className="zoomable-image-inner"

BIN
docs/static/img/document_processor.png vendored Normal file
View file

Binary file not shown.
Side by side

After

Width:  |  Height:  |  Size: 197 KiB

BIN
docs/static/img/document_processor_code.png vendored Normal file
View file

Binary file not shown.
Side by side

After

Width:  |  Height:  |  Size: 407 KiB

BIN
docs/static/img/flow_runner.png vendored Normal file
View file

Binary file not shown.
Side by side

After

Width:  |  Height:  |  Size: 129 KiB

BIN
docs/static/img/flow_runner_code.png vendored Normal file
View file

Binary file not shown.
Side by side

After

Width:  |  Height:  |  Size: 379 KiB

BIN
docs/static/img/widget-code.png vendored Normal file
View file

Binary file not shown.
Side by side

After

Width:  |  Height:  |  Size: 787 KiB

BIN
docs/static/img/widget-sidebar.png vendored Normal file
View file

Binary file not shown.
Side by side

After

Width:  |  Height:  |  Size: 129 KiB

BIN
docs/static/videos/langflow_widget.mp4 vendored Normal file
View file

Binary file not shown.

831
poetry.lock generated
View file

File diff suppressed because it is too large Load diff

18
pyproject.toml
View file

@ -1,6 +1,6 @@
[tool.poetry]
name = "langflow"
version = "0.3.3"
version = "0.4.0"
description = "A Python package with a built-in web application"
authors = ["Logspace <contact@logspace.ai>"]
maintainers = [
@ -33,13 +33,13 @@ google-search-results = "^2.4.1"
google-api-python-client = "^2.79.0"
typer = "^0.9.0"
gunicorn = "^21.1.0"
langchain = "^0.0.240"
langchain = "^0.0.249"
openai = "^0.27.8"
pandas = "^2.0.0"
chromadb = "^0.3.21"
huggingface-hub = "^0.15.0"
huggingface-hub = { version = "^0.16.0", extras = ["inference"] }
rich = "^13.4.2"
llama-cpp-python = "~0.1.0"
llama-cpp-python = { version = "~0.1.0", optional = true }
networkx = "^3.1"
unstructured = "^0.7.0"
pypdf = "^3.11.0"
@ -56,8 +56,8 @@ qdrant-client = "^1.3.0"
websockets = "^10.3"
weaviate-client = "^3.21.0"
jina = "3.15.2"
sentence-transformers = "^2.2.2"
ctransformers = "^0.2.10"
sentence-transformers = { version = "^2.2.2", optional = true }
ctransformers = { version = "^0.2.10", optional = true }
cohere = "^4.11.0"
python-multipart = "^0.0.6"
sqlmodel = "^0.0.8"
@ -75,8 +75,9 @@ certifi = "^2023.5.7"
google-cloud-aiplatform = "^1.26.1"
psycopg = "^3.1.9"
psycopg-binary = "^3.1.9"
fastavro = "^1.8.0"
[tool.poetry.dev-dependencies]
[tool.poetry.group.dev.dependencies]
black = "^23.1.0"
ipykernel = "^6.21.2"
mypy = "^1.1.1"
@ -94,6 +95,9 @@ types-pyyaml = "^6.0.12.8"
[tool.poetry.extras]
deploy = ["langchain-serve"]
local = ["llama-cpp-python", "sentence-transformers", "ctransformers"]
all = ["deploy", "local"]
[tool.pytest.ini_options]
minversion = "6.0"

8
src/backend/langflow/__init__.py
View file

@ -1,6 +1,7 @@
from importlib import metadata
from langflow.cache import cache_manager # noqa: E402
from langflow.processing.process import load_flow_from_json # noqa: E402
from langflow.cache import cache_manager
from langflow.processing.process import load_flow_from_json
from langflow.interface.custom.custom_component import CustomComponent
try:
__version__ = metadata.version(__package__)
@ -9,5 +10,4 @@ except metadata.PackageNotFoundError:
__version__ = ""
del metadata # optional, avoids polluting the results of dir(__package__)
__all__ = ["load_flow_from_json", "cache_manager"]
__all__ = ["load_flow_from_json", "cache_manager", "CustomComponent"]

28
src/backend/langflow/__main__.py
View file

@ -2,7 +2,8 @@ import os
import sys
import time
import httpx
from multiprocess import Process, cpu_count # type: ignore
from langflow.utils.util import get_number_of_workers
from multiprocess import Process # type: ignore
import platform
from pathlib import Path
from typing import Optional
@ -20,18 +21,13 @@ from dotenv import load_dotenv
app = typer.Typer()
def get_number_of_workers(workers=None):
if workers == -1:
workers = (cpu_count() * 2) + 1
return workers
def update_settings(
config: str,
cache: str,
dev: bool = False,
database_url: Optional[str] = None,
remove_api_keys: bool = False,
components_path: Optional[Path] = None,
):
"""Update the settings from a config file."""
@ -39,13 +35,19 @@ def update_settings(
database_url = database_url or os.getenv("langflow_database_url")
if config:
logger.debug(f"Loading settings from {config}")
settings.update_from_yaml(config, dev=dev)
if database_url:
settings.update_settings(database_url=database_url)
if remove_api_keys:
logger.debug(f"Setting remove_api_keys to {remove_api_keys}")
settings.update_settings(remove_api_keys=remove_api_keys)
if cache:
logger.debug(f"Setting cache to {cache}")
settings.update_settings(cache=cache)
if components_path:
logger.debug(f"Adding component path {components_path}")
settings.update_settings(components_path=components_path)
def load_params():
@ -120,10 +122,15 @@ def serve(
"127.0.0.1", help="Host to bind the server to.", envvar="LANGFLOW_HOST"
),
workers: int = typer.Option(
1, help="Number of worker processes.", envvar="LANGFLOW_WORKERS"
2, help="Number of worker processes.", envvar="LANGFLOW_WORKERS"
),
timeout: int = typer.Option(60, help="Worker timeout in seconds."),
timeout: int = typer.Option(300, help="Worker timeout in seconds."),
port: int = typer.Option(7860, help="Port to listen on.", envvar="LANGFLOW_PORT"),
components_path: Optional[Path] = typer.Option(
Path(__file__).parent / "components",
help="Path to the directory containing custom components.",
envvar="LANGFLOW_COMPONENTS_PATH",
),
config: str = typer.Option("config.yaml", help="Path to the configuration file."),
# .env file param
env_file: Path = typer.Option(
@ -181,6 +188,7 @@ def serve(
database_url=database_url,
remove_api_keys=remove_api_keys,
cache=cache,
components_path=components_path,
)
# create path object if path is provided
static_files_dir: Optional[Path] = Path(path) if path else None
@ -298,7 +306,7 @@ def run_langflow(host, port, log_level, options, app):
Run Langflow server on localhost
"""
try:
if platform.system() in ["Darwin", "Windows"]:
if platform.system() in ["Windows"]:
# Run using uvicorn on MacOS and Windows
# Windows doesn't support gunicorn
# MacOS requires an env variable to be set to use gunicorn

2
src/backend/langflow/api/router.py
View file

@ -6,6 +6,7 @@ from langflow.api.v1 import (
validate_router,
flows_router,
flow_styles_router,
component_router,
)
router = APIRouter(
@ -14,5 +15,6 @@ router = APIRouter(
router.include_router(chat_router)
router.include_router(endpoints_router)
router.include_router(validate_router)
router.include_router(component_router)
router.include_router(flows_router)
router.include_router(flow_styles_router)

9
src/backend/langflow/api/utils.py
View file

@ -57,3 +57,12 @@ def build_input_keys_response(langchain_object, artifacts):
input_keys_response["template"] = langchain_object.prompt.template
return input_keys_response
def merge_nested_dicts(dict1, dict2):
for key, value in dict2.items():
if isinstance(value, dict) and isinstance(dict1.get(key), dict):
dict1[key] = merge_nested_dicts(dict1[key], value)
else:
dict1[key] = value
return dict1

2
src/backend/langflow/api/v1/__init__.py
View file

@ -3,10 +3,12 @@ from langflow.api.v1.validate import router as validate_router
from langflow.api.v1.chat import router as chat_router
from langflow.api.v1.flows import router as flows_router
from langflow.api.v1.flow_styles import router as flow_styles_router
from langflow.api.v1.components import router as component_router
__all__ = [
"chat_router",
"endpoints_router",
"component_router",
"validate_router",
"flows_router",
"flow_styles_router",

4
src/backend/langflow/api/v1/callback.py
View file

@ -91,8 +91,8 @@ class AsyncStreamingLLMCallbackHandler(AsyncCallbackHandler):
# This is to emulate the stream of tokens
for resp in resps:
await self.websocket.send_json(resp.dict())
except Exception as e:
logger.error(e)
except Exception as exc:
logger.error(f"Error sending response: {exc}")
async def on_tool_error(
self, error: Union[Exception, KeyboardInterrupt], **kwargs: Any

8
src/backend/langflow/api/v1/chat.py
View file

@ -26,7 +26,7 @@ async def chat(client_id: str, websocket: WebSocket):
message = "Please, build the flow before sending messages"
await websocket.close(code=status.WS_1011_INTERNAL_ERROR, reason=message)
except WebSocketException as exc:
logger.error(exc)
logger.error(f"Websocket error: {exc}")
await websocket.close(code=status.WS_1011_INTERNAL_ERROR, reason=str(exc))
@ -56,7 +56,7 @@ async def init_build(graph_data: dict, flow_id: str):
return InitResponse(flowId=flow_id)
except Exception as exc:
logger.error(exc)
logger.error(f"Error initializing build: {exc}")
return HTTPException(status_code=500, detail=str(exc))
@ -74,7 +74,7 @@ async def build_status(flow_id: str):
)
except Exception as exc:
logger.error(exc)
logger.error(f"Error checking build status: {exc}")
return HTTPException(status_code=500, detail=str(exc))
@ -177,5 +177,5 @@ async def stream_build(flow_id: str):
try:
return StreamingResponse(event_stream(flow_id), media_type="text/event-stream")
except Exception as exc:
logger.error(exc)
logger.error(f"Error streaming build: {exc}")
raise HTTPException(status_code=500, detail=str(exc))

77
src/backend/langflow/api/v1/components.py Normal file
View file

@ -0,0 +1,77 @@
from datetime import timezone
from typing import List
from uuid import UUID
from langflow.database.models.component import Component, ComponentModel
from langflow.database.base import get_session
from sqlmodel import Session, select
from fastapi import APIRouter, Depends, HTTPException
from sqlalchemy.exc import IntegrityError
from datetime import datetime
COMPONENT_NOT_FOUND = "Component not found"
COMPONENT_ALREADY_EXISTS = "A component with the same id already exists."
COMPONENT_DELETED = "Component deleted"
router = APIRouter(prefix="/components", tags=["Components"])
@router.post("/", response_model=Component)
def create_component(component: ComponentModel, db: Session = Depends(get_session)):
db_component = Component(**component.dict())
try:
db.add(db_component)
db.commit()
db.refresh(db_component)
except IntegrityError as e:
db.rollback()
raise HTTPException(
status_code=400,
detail=COMPONENT_ALREADY_EXISTS,
) from e
return db_component
@router.get("/{component_id}", response_model=Component)
def read_component(component_id: UUID, db: Session = Depends(get_session)):
if component := db.get(Component, component_id):
return component
else:
raise HTTPException(status_code=404, detail=COMPONENT_NOT_FOUND)
@router.get("/", response_model=List[Component])
def read_components(skip: int = 0, limit: int = 50, db: Session = Depends(get_session)):
query = select(Component)
query = query.offset(skip).limit(limit)
return db.execute(query).fetchall()
@router.patch("/{component_id}", response_model=Component)
def update_component(
component_id: UUID, component: ComponentModel, db: Session = Depends(get_session)
):
db_component = db.get(Component, component_id)
if not db_component:
raise HTTPException(status_code=404, detail=COMPONENT_NOT_FOUND)
component_data = component.dict(exclude_unset=True)
for key, value in component_data.items():
setattr(db_component, key, value)
db_component.update_at = datetime.now(timezone.utc)
db.commit()
db.refresh(db_component)
return db_component
@router.delete("/{component_id}")
def delete_component(component_id: UUID, db: Session = Depends(get_session)):
component = db.get(Component, component_id)
if not component:
raise HTTPException(status_code=404, detail=COMPONENT_NOT_FOUND)
db.delete(component)
db.commit()
return {"detail": COMPONENT_DELETED}

84
src/backend/langflow/api/v1/endpoints.py
View file

@ -1,17 +1,34 @@
from typing import Optional
from http import HTTPStatus
from typing import Annotated, Optional
from langflow.cache.utils import save_uploaded_file
from langflow.database.models.flow import Flow
from langflow.processing.process import process_graph_cached, process_tweaks
from langflow.utils.logger import logger
from langflow.settings import settings
from fastapi import APIRouter, Depends, HTTPException, UploadFile
from fastapi import APIRouter, Depends, HTTPException, UploadFile, Body
from langflow.interface.custom.custom_component import CustomComponent
from langflow.interface.custom.directory_reader import (
CustomComponentPathValueError,
)
from langflow.api.v1.schemas import (
ProcessResponse,
UploadFileResponse,
CustomComponentCode,
)
from langflow.api.utils import merge_nested_dicts
from langflow.interface.types import (
build_langchain_types_dict,
build_langchain_template_custom_component,
build_langchain_custom_component_list_from_path,
)
from langflow.interface.types import langchain_types_dict
from langflow.database.base import get_session
from sqlmodel import Session
@ -21,7 +38,47 @@ router = APIRouter(tags=["Base"])
@router.get("/all")
def get_all():
return langchain_types_dict
native_components = build_langchain_types_dict()
# custom_components is a list of dicts
# need to merge all the keys into one dict
custom_components_from_file = {}
if settings.components_path:
custom_component_dicts = [
build_langchain_custom_component_list_from_path(str(path))
for path in settings.components_path
]
for custom_component_dict in custom_component_dicts:
custom_components_from_file = merge_nested_dicts(
custom_components_from_file, custom_component_dict
)
return merge_nested_dicts(native_components, custom_components_from_file)
@router.get("/load_custom_component_from_path")
def get_load_custom_component_from_path(path: str):
try:
data = build_langchain_custom_component_list_from_path(path)
except CustomComponentPathValueError as err:
raise HTTPException(
status_code=400,
detail={"error": type(err).__name__, "traceback": str(err)},
) from err
return data
@router.get("/load_custom_component_from_path_TEST")
def get_load_custom_component_from_path_test(path: str):
from langflow.interface.custom.directory_reader import (
DirectoryReader,
)
reader = DirectoryReader(path, False)
file_list = reader.get_files()
data = reader.build_component_menu_list(file_list)
return reader.filter_loaded_components(data, True)
# For backwards compatibility we will keep the old endpoint
@ -31,6 +88,7 @@ async def process_flow(
flow_id: str,
inputs: Optional[dict] = None,
tweaks: Optional[dict] = None,
clear_cache: Annotated[bool, Body(embed=True)] = False, # noqa: F821
session: Session = Depends(get_session),
):
"""
@ -50,7 +108,7 @@ async def process_flow(
graph_data = process_tweaks(graph_data, tweaks)
except Exception as exc:
logger.error(f"Error processing tweaks: {exc}")
response = process_graph_cached(graph_data, inputs)
response = process_graph_cached(graph_data, inputs, clear_cache)
return ProcessResponse(
result=response,
)
@ -60,7 +118,11 @@ async def process_flow(
raise HTTPException(status_code=500, detail=str(e)) from e
@router.post("/upload/{flow_id}", response_model=UploadFileResponse, status_code=201)
@router.post(
"/upload/{flow_id}",
response_model=UploadFileResponse,
status_code=HTTPStatus.CREATED,
)
async def create_upload_file(file: UploadFile, flow_id: str):
# Cache file
try:
@ -81,3 +143,13 @@ def get_version():
from langflow import __version__
return {"version": __version__}
@router.post("/custom_component", status_code=HTTPStatus.OK)
async def custom_component(
raw_code: CustomComponentCode,
):
extractor = CustomComponent(code=raw_code.code)
extractor.is_check_valid()
return build_langchain_template_custom_component(extractor)

17
src/backend/langflow/api/v1/schemas.py
View file

@ -116,3 +116,20 @@ class StreamData(BaseModel):
def __str__(self) -> str:
return f"event: {self.event}\ndata: {json.dumps(self.data)}\n\n"
class CustomComponentCode(BaseModel):
code: str
class CustomComponentResponseError(BaseModel):
detail: str
traceback: str
class ComponentListCreate(BaseModel):
flows: List[FlowCreate]
class ComponentListRead(BaseModel):
flows: List[FlowRead]

12
src/backend/langflow/chat/manager.py
View file

@ -111,7 +111,7 @@ class ChatManager:
# This is to catch the following error:
# Unexpected ASGI message 'websocket.close', after sending 'websocket.close'
if "after sending" in str(exc):
logger.error(exc)
logger.error(f"Error closing connection: {exc}")
async def process_message(
self, client_id: str, payload: Dict, langchain_object: Any
@ -197,13 +197,13 @@ class ChatManager:
langchain_object = self.in_memory_cache.get(client_id)
await self.process_message(client_id, payload, langchain_object)
except Exception as e:
except Exception as exc:
# Handle any exceptions that might occur
logger.error(e)
logger.error(f"Error handling websocket: {exc}")
await self.close_connection(
client_id=client_id,
code=status.WS_1011_INTERNAL_ERROR,
reason=str(e)[:120],
reason=str(exc)[:120],
)
finally:
try:
@ -212,6 +212,6 @@ class ChatManager:
code=status.WS_1000_NORMAL_CLOSURE,
reason="Client disconnected",
)
except Exception as e:
logger.error(e)
except Exception as exc:
logger.error(f"Error closing connection: {exc}")
self.disconnect(client_id)

5
src/backend/langflow/config.yaml
View file

@ -153,6 +153,8 @@ memories:
documentation: "https://python.langchain.com/docs/modules/memory/how_to/vectorstore_retriever_memory"
MongoDBChatMessageHistory:
documentation: "https://python.langchain.com/docs/modules/memory/integrations/mongodb_chat_message_history"
MotorheadMemory:
documentation: "https://python.langchain.com/docs/integrations/memory/motorhead_memory"
prompts:
ChatMessagePromptTemplate:
documentation: "https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/msg_prompt_templates"
@ -290,3 +292,6 @@ output_parsers:
documentation: "https://python.langchain.com/docs/modules/model_io/output_parsers/structured"
ResponseSchema:
documentation: "https://python.langchain.com/docs/modules/model_io/output_parsers/structured"
custom_components:
CustomComponent:
documentation: ""

3
src/backend/langflow/custom/customs.py
View file

@ -31,6 +31,9 @@ CUSTOM_NODES = {
"MidJourneyPromptChain": frontend_node.chains.MidJourneyPromptChainNode(),
"load_qa_chain": frontend_node.chains.CombineDocsChainNode(),
},
"custom_components": {
"CustomComponent": frontend_node.custom_components.CustomComponentFrontendNode(),
},
}

18
src/backend/langflow/database/base.py
View file

@ -1,3 +1,4 @@
from contextlib import contextmanager
from langflow.settings import settings
from sqlmodel import SQLModel, Session, create_engine
from langflow.utils.logger import logger
@ -32,6 +33,19 @@ def create_db_and_tables():
logger.debug("Database and tables created successfully")
def get_session():
with Session(engine) as session:
@contextmanager
def session_getter():
try:
session = Session(engine)
yield session
except Exception as e:
print("Session rollback because of exception:", e)
session.rollback()
raise
finally:
session.close()
def get_session():
with session_getter() as session:
yield session

29
src/backend/langflow/database/models/component.py Normal file
View file

@ -0,0 +1,29 @@
from langflow.database.models.base import SQLModelSerializable, SQLModel
from sqlmodel import Field
from typing import Optional
from datetime import datetime
import uuid
class Component(SQLModelSerializable, table=True):
id: uuid.UUID = Field(default_factory=uuid.uuid4, primary_key=True)
frontend_node_id: uuid.UUID = Field(index=True)
name: str = Field(index=True)
description: Optional[str] = Field(default=None)
python_code: Optional[str] = Field(default=None)
return_type: Optional[str] = Field(default=None)
is_disabled: bool = Field(default=False)
is_read_only: bool = Field(default=False)
create_at: datetime = Field(default_factory=datetime.utcnow)
update_at: datetime = Field(default_factory=datetime.utcnow)
class ComponentModel(SQLModel):
id: uuid.UUID = Field(default_factory=uuid.uuid4)
frontend_node_id: uuid.UUID = Field(default=uuid.uuid4())
name: str = Field(default="")
description: Optional[str] = None
python_code: Optional[str] = None
return_type: Optional[str] = None
is_disabled: bool = False
is_read_only: bool = False

2
src/backend/langflow/graph/graph/base.py
View file

@ -77,6 +77,8 @@ class Graph:
def _validate_nodes(self) -> None:
"""Check that all nodes have edges"""
if len(self.nodes) == 1:
return
for node in self.nodes:
if not self._validate_node(node):
raise ValueError(

3
src/backend/langflow/graph/graph/constants.py
View file

@ -14,7 +14,7 @@ from langflow.interface.vector_store.base import vectorstore_creator
from langflow.interface.wrappers.base import wrapper_creator
from langflow.interface.output_parsers.base import output_parser_creator
from langflow.interface.retrievers.base import retriever_creator
from langflow.interface.custom.base import custom_component_creator
from typing import Dict, Type
@ -32,5 +32,6 @@ VERTEX_TYPE_MAP: Dict[str, Type[Vertex]] = {
**{t: types.DocumentLoaderVertex for t in documentloader_creator.to_list()},
**{t: types.TextSplitterVertex for t in textsplitter_creator.to_list()},
**{t: types.OutputParserVertex for t in output_parser_creator.to_list()},
**{t: types.CustomComponentVertex for t in custom_component_creator.to_list()},
**{t: types.RetrieverVertex for t in retriever_creator.to_list()},
}

9
src/backend/langflow/graph/vertex/types.py
View file

@ -239,3 +239,12 @@ class PromptVertex(Vertex):
class OutputParserVertex(Vertex):
def __init__(self, data: Dict):
super().__init__(data, base_type="output_parsers")
class CustomComponentVertex(Vertex):
def __init__(self, data: Dict):
super().__init__(data, base_type="custom_components")
def _built_object_repr(self):
if self.artifacts and "repr" in self.artifacts:
return self.artifacts["repr"] or super()._built_object_repr()

2
src/backend/langflow/interface/base.py
View file

@ -34,7 +34,7 @@ class LangChainTypeCreator(BaseModel, ABC):
for name, value_dict in type_settings.items()
}
except AttributeError as exc:
logger.error(exc)
logger.error(f"Error getting settings for {self.type_name}: {exc}")
self.name_docs_dict = {}
return self.name_docs_dict

4
src/backend/langflow/interface/custom/__init__.py Normal file
View file

@ -0,0 +1,4 @@
from langflow.interface.custom.base import CustomComponentCreator
from langflow.interface.custom.custom_component import CustomComponent
__all__ = ["CustomComponentCreator", "CustomComponent"]

48
src/backend/langflow/interface/custom/base.py Normal file
View file

@ -0,0 +1,48 @@
from typing import Any, Dict, List, Optional, Type
from langflow.interface.base import LangChainTypeCreator
# from langflow.interface.custom.custom import CustomComponent
from langflow.interface.custom.custom_component import CustomComponent
from langflow.template.frontend_node.custom_components import (
CustomComponentFrontendNode,
)
from langflow.utils.logger import logger
# Assuming necessary imports for Field, Template, and FrontendNode classes
class CustomComponentCreator(LangChainTypeCreator):
type_name: str = "custom_components"
@property
def frontend_node_class(self) -> Type[CustomComponentFrontendNode]:
return CustomComponentFrontendNode
@property
def type_to_loader_dict(self) -> Dict:
if self.type_dict is None:
self.type_dict: dict[str, Any] = {
"CustomComponent": CustomComponent,
}
return self.type_dict
def get_signature(self, name: str) -> Optional[Dict]:
from langflow.custom.customs import get_custom_nodes
try:
if name in get_custom_nodes(self.type_name).keys():
return get_custom_nodes(self.type_name)[name]
except ValueError as exc:
raise ValueError(f"CustomComponent {name} not found: {exc}") from exc
except AttributeError as exc:
logger.error(f"CustomComponent {name} not loaded: {exc}")
return None
return None
def to_list(self) -> List[str]:
return list(self.type_to_loader_dict.keys())
custom_component_creator = CustomComponentCreator()

272
src/backend/langflow/interface/custom/code_parser.py Normal file
View file

@ -0,0 +1,272 @@
import ast
import inspect
import traceback
from typing import Dict, Any, List, Type, Union
from fastapi import HTTPException
from langflow.interface.custom.schema import CallableCodeDetails, ClassCodeDetails
class CodeSyntaxError(HTTPException):
pass
class CodeParser:
"""
A parser for Python source code, extracting code details.
"""
def __init__(self, code: Union[str, Type]) -> None:
"""
Initializes the parser with the provided code.
"""
if isinstance(code, type):
if not inspect.isclass(code):
raise ValueError("The provided code must be a class.")
# If the code is a class, get its source code
code = inspect.getsource(code)
self.code = code
self.data: Dict[str, Any] = {
"imports": [],
"functions": [],
"classes": [],
"global_vars": [],
}
self.handlers = {
ast.Import: self.parse_imports,
ast.ImportFrom: self.parse_imports,
ast.FunctionDef: self.parse_functions,
ast.ClassDef: self.parse_classes,
ast.Assign: self.parse_global_vars,
}
def __get_tree(self):
"""
Parses the provided code to validate its syntax.
It tries to parse the code into an abstract syntax tree (AST).
"""
try:
tree = ast.parse(self.code)
except SyntaxError as err:
raise CodeSyntaxError(
status_code=400,
detail={"error": err.msg, "traceback": traceback.format_exc()},
) from err
return tree
def parse_node(self, node: Union[ast.stmt, ast.AST]) -> None:
"""
Parses an AST node and updates the data
dictionary with the relevant information.
"""
if handler := self.handlers.get(type(node)): # type: ignore
handler(node) # type: ignore
def parse_imports(self, node: Union[ast.Import, ast.ImportFrom]) -> None:
"""
Extracts "imports" from the code.
"""
if isinstance(node, ast.Import):
for alias in node.names:
self.data["imports"].append(alias.name)
elif isinstance(node, ast.ImportFrom):
for alias in node.names:
self.data["imports"].append((node.module, alias.name))
def parse_functions(self, node: ast.FunctionDef) -> None:
"""
Extracts "functions" from the code.
"""
self.data["functions"].append(self.parse_callable_details(node))
def parse_arg(self, arg, default):
"""
Parses an argument and its default value.
"""
arg_dict = {"name": arg.arg, "default": default}
if arg.annotation:
arg_dict["type"] = ast.unparse(arg.annotation)
return arg_dict
def parse_callable_details(self, node: ast.FunctionDef) -> Dict[str, Any]:
"""
Extracts details from a single function or method node.
"""
func = CallableCodeDetails(
name=node.name,
doc=ast.get_docstring(node),
args=[],
body=[],
return_type=ast.unparse(node.returns) if node.returns else None,
)
func.args = self.parse_function_args(node)
func.body = self.parse_function_body(node)
return func.dict()
def parse_function_args(self, node: ast.FunctionDef) -> List[Dict[str, Any]]:
"""
Parses the arguments of a function or method node.
"""
args = []
args += self.parse_positional_args(node)
args += self.parse_varargs(node)
args += self.parse_keyword_args(node)
args += self.parse_kwargs(node)
return args
def parse_positional_args(self, node: ast.FunctionDef) -> List[Dict[str, Any]]:
"""
Parses the positional arguments of a function or method node.
"""
num_args = len(node.args.args)
num_defaults = len(node.args.defaults)
num_missing_defaults = num_args - num_defaults
missing_defaults = [None] * num_missing_defaults
default_values = [
ast.unparse(default).strip("'") if default else None
for default in node.args.defaults
]
# Now check all default values to see if there
# are any "None" values in the middle
default_values = [
None if value == "None" else value for value in default_values
]
defaults = missing_defaults + default_values
args = [
self.parse_arg(arg, default)
for arg, default in zip(node.args.args, defaults)
]
return args
def parse_varargs(self, node: ast.FunctionDef) -> List[Dict[str, Any]]:
"""
Parses the *args argument of a function or method node.
"""
args = []
if node.args.vararg:
args.append(self.parse_arg(node.args.vararg, None))
return args
def parse_keyword_args(self, node: ast.FunctionDef) -> List[Dict[str, Any]]:
"""
Parses the keyword-only arguments of a function or method node.
"""
kw_defaults = [None] * (
len(node.args.kwonlyargs) - len(node.args.kw_defaults)
) + [
ast.unparse(default) if default else None
for default in node.args.kw_defaults
]
args = [
self.parse_arg(arg, default)
for arg, default in zip(node.args.kwonlyargs, kw_defaults)
]
return args
def parse_kwargs(self, node: ast.FunctionDef) -> List[Dict[str, Any]]:
"""
Parses the **kwargs argument of a function or method node.
"""
args = []
if node.args.kwarg:
args.append(self.parse_arg(node.args.kwarg, None))
return args
def parse_function_body(self, node: ast.FunctionDef) -> List[str]:
"""
Parses the body of a function or method node.
"""
return [ast.unparse(line) for line in node.body]
def parse_assign(self, stmt):
"""
Parses an Assign statement and returns a dictionary
with the target's name and value.
"""
for target in stmt.targets:
if isinstance(target, ast.Name):
return {"name": target.id, "value": ast.unparse(stmt.value)}
def parse_ann_assign(self, stmt):
"""
Parses an AnnAssign statement and returns a dictionary
with the target's name, value, and annotation.
"""
if isinstance(stmt.target, ast.Name):
return {
"name": stmt.target.id,
"value": ast.unparse(stmt.value) if stmt.value else None,
"annotation": ast.unparse(stmt.annotation),
}
def parse_function_def(self, stmt):
"""
Parses a FunctionDef statement and returns the parsed
method and a boolean indicating if it's an __init__ method.
"""
method = self.parse_callable_details(stmt)
return (method, True) if stmt.name == "__init__" else (method, False)
def parse_classes(self, node: ast.ClassDef) -> None:
"""
Extracts "classes" from the code, including inheritance and init methods.
"""
class_details = ClassCodeDetails(
name=node.name,
doc=ast.get_docstring(node),
bases=[ast.unparse(base) for base in node.bases],
attributes=[],
methods=[],
init=None,
)
for stmt in node.body:
if isinstance(stmt, ast.Assign):
if attr := self.parse_assign(stmt):
class_details.attributes.append(attr)
elif isinstance(stmt, ast.AnnAssign):
if attr := self.parse_ann_assign(stmt):
class_details.attributes.append(attr)
elif isinstance(stmt, ast.FunctionDef):
method, is_init = self.parse_function_def(stmt)
if is_init:
class_details.init = method
else:
class_details.methods.append(method)
self.data["classes"].append(class_details.dict())
def parse_global_vars(self, node: ast.Assign) -> None:
"""
Extracts global variables from the code.
"""
global_var = {
"targets": [
t.id if hasattr(t, "id") else ast.dump(t) for t in node.targets
],
"value": ast.unparse(node.value),
}
self.data["global_vars"].append(global_var)
def parse_code(self) -> Dict[str, Any]:
"""
Runs all parsing operations and returns the resulting data.
"""
tree = self.__get_tree()
for node in ast.walk(tree):
self.parse_node(node)
return self.data

72
src/backend/langflow/interface/custom/component.py Normal file
View file

@ -0,0 +1,72 @@
import ast
from typing import Optional
from pydantic import BaseModel
from fastapi import HTTPException
from langflow.utils import validate
from langflow.interface.custom.code_parser import CodeParser
class ComponentCodeNullError(HTTPException):
pass
class ComponentFunctionEntrypointNameNullError(HTTPException):
pass
class Component(BaseModel):
ERROR_CODE_NULL = "Python code must be provided."
ERROR_FUNCTION_ENTRYPOINT_NAME_NULL = (
"The name of the entrypoint function must be provided."
)
code: Optional[str]
function_entrypoint_name = "build"
field_config: dict = {}
def __init__(self, **data):
super().__init__(**data)
def get_code_tree(self, code: str):
parser = CodeParser(code)
return parser.parse_code()
def get_function(self):
if not self.code:
raise ComponentCodeNullError(
status_code=400,
detail={"error": self.ERROR_CODE_NULL, "traceback": ""},
)
if not self.function_entrypoint_name:
raise ComponentFunctionEntrypointNameNullError(
status_code=400,
detail={
"error": self.ERROR_FUNCTION_ENTRYPOINT_NAME_NULL,
"traceback": "",
},
)
return validate.create_function(self.code, self.function_entrypoint_name)
def build_template_config(self, attributes) -> dict:
template_config = {}
for item in attributes:
item_name = item.get("name")
if item_value := item.get("value"):
if "display_name" in item_name:
template_config["display_name"] = ast.literal_eval(item_value)
elif "description" in item_name:
template_config["description"] = ast.literal_eval(item_value)
elif "field_config" in item_name:
template_config["field_config"] = ast.literal_eval(item_value)
return template_config
def build(self):
raise NotImplementedError

58
src/backend/langflow/interface/custom/constants.py Normal file
View file

@ -0,0 +1,58 @@
from langchain import PromptTemplate
from langchain.chains.base import Chain
from langchain.document_loaders.base import BaseLoader
from langchain.embeddings.base import Embeddings
from langchain.llms.base import BaseLLM
from langchain.schema import BaseRetriever, Document
from langchain.text_splitter import TextSplitter
from langchain.tools import Tool
from langchain.vectorstores.base import VectorStore
LANGCHAIN_BASE_TYPES = {
"Chain": Chain,
"Tool": Tool,
"BaseLLM": BaseLLM,
"PromptTemplate": PromptTemplate,
"BaseLoader": BaseLoader,
"Document": Document,
"TextSplitter": TextSplitter,
"VectorStore": VectorStore,
"Embeddings": Embeddings,
"BaseRetriever": BaseRetriever,
}
# Langchain base types plus Python base types
CUSTOM_COMPONENT_SUPPORTED_TYPES = {
**LANGCHAIN_BASE_TYPES,
"str": str,
"int": int,
"float": float,
"bool": bool,
"list": list,
"dict": dict,
}
DEFAULT_CUSTOM_COMPONENT_CODE = """from langflow import CustomComponent
from langchain.llms.base import BaseLLM
from langchain.chains import LLMChain
from langchain import PromptTemplate
from langchain.schema import Document
import requests
class YourComponent(CustomComponent):
display_name: str = "Custom Component"
description: str = "Create any custom component you want!"
def build_config(self):
return { "url": { "multiline": True, "required": True } }
def build(self, url: str, llm: BaseLLM, prompt: PromptTemplate) -> Document:
response = requests.get(url)
chain = LLMChain(llm=llm, prompt=prompt)
result = chain.run(response.text[:300])
return Document(page_content=str(result))
"""

194
src/backend/langflow/interface/custom/custom_component.py Normal file
View file

@ -0,0 +1,194 @@
from typing import Any, Callable, List, Optional
from fastapi import HTTPException
from langflow.interface.custom.constants import CUSTOM_COMPONENT_SUPPORTED_TYPES
from langflow.interface.custom.component import Component
from langflow.interface.custom.directory_reader import DirectoryReader
from langflow.utils import validate
from langflow.database.base import session_getter
from langflow.database.models.flow import Flow
from pydantic import Extra
class CustomComponent(Component, extra=Extra.allow):
code: Optional[str]
field_config: dict = {}
code_class_base_inheritance = "CustomComponent"
function_entrypoint_name = "build"
function: Optional[Callable] = None
return_type_valid_list = list(CUSTOM_COMPONENT_SUPPORTED_TYPES.keys())
repr_value: Optional[str] = ""
def __init__(self, **data):
super().__init__(**data)
def custom_repr(self):
return str(self.repr_value)
def build_config(self):
return self.field_config
def _class_template_validation(self, code: str):
TYPE_HINT_LIST = ["Optional", "Prompt", "PromptTemplate", "LLMChain"]
if not code:
raise HTTPException(
status_code=400,
detail={
"error": self.ERROR_CODE_NULL,
"traceback": "",
},
)
reader = DirectoryReader("", False)
for type_hint in TYPE_HINT_LIST:
if reader.is_type_hint_used_but_not_imported(type_hint, code):
error_detail = {
"error": "Type hint Error",
"traceback": f"Type hint '{type_hint}' is used but not imported in the code.",
}
raise HTTPException(status_code=400, detail=error_detail)
def is_check_valid(self) -> bool:
return self._class_template_validation(self.code) if self.code else False
def get_code_tree(self, code: str):
return super().get_code_tree(code)
@property
def get_function_entrypoint_args(self) -> str:
if not self.code:
return ""
tree = self.get_code_tree(self.code)
component_classes = [
cls
for cls in tree["classes"]
if self.code_class_base_inheritance in cls["bases"]
]
if not component_classes:
return ""
# Assume the first Component class is the one we're interested in
component_class = component_classes[0]
build_methods = [
method
for method in component_class["methods"]
if method["name"] == self.function_entrypoint_name
]
if not build_methods:
return ""
build_method = build_methods[0]
return build_method["args"]
@property
def get_function_entrypoint_return_type(self) -> str:
if not self.code:
return ""
tree = self.get_code_tree(self.code)
component_classes = [
cls
for cls in tree["classes"]
if self.code_class_base_inheritance in cls["bases"]
]
if not component_classes:
return ""
# Assume the first Component class is the one we're interested in
component_class = component_classes[0]
build_methods = [
method
for method in component_class["methods"]
if method["name"] == self.function_entrypoint_name
]
if not build_methods:
return ""
build_method = build_methods[0]
return build_method["return_type"]
@property
def get_main_class_name(self):
tree = self.get_code_tree(self.code)
base_name = self.code_class_base_inheritance
method_name = self.function_entrypoint_name
classes = []
for item in tree.get("classes"):
if base_name in item["bases"]:
method_names = [method["name"] for method in item["methods"]]
if method_name in method_names:
classes.append(item["name"])
# Get just the first item
return next(iter(classes), "")
@property
def build_template_config(self):
tree = self.get_code_tree(self.code)
attributes = [
main_class["attributes"]
for main_class in tree.get("classes")
if main_class["name"] == self.get_main_class_name
]
# Get just the first item
attributes = next(iter(attributes), [])
return super().build_template_config(attributes)
@property
def get_function(self):
return validate.create_function(self.code, self.function_entrypoint_name)
def load_flow(self, flow_id: str, tweaks: Optional[dict] = None) -> Any:
from langflow.processing.process import build_sorted_vertices_with_caching
from langflow.processing.process import process_tweaks
with session_getter() as session:
graph_data = flow.data if (flow := session.get(Flow, flow_id)) else None
if not graph_data:
raise ValueError(f"Flow {flow_id} not found")
if tweaks:
graph_data = process_tweaks(graph_data=graph_data, tweaks=tweaks)
return build_sorted_vertices_with_caching(graph_data)
def list_flows(self, *, get_session: Optional[Callable] = None) -> List[Flow]:
get_session = get_session or session_getter
with get_session() as session:
flows = session.query(Flow).all()
return flows
def get_flow(
self,
*,
flow_name: Optional[str] = None,
flow_id: Optional[str] = None,
tweaks: Optional[dict] = None,
get_session: Optional[Callable] = None,
) -> Flow:
get_session = get_session or session_getter
with get_session() as session:
if flow_id:
flow = session.query(Flow).get(flow_id)
elif flow_name:
flow = session.query(Flow).filter(Flow.name == flow_name).first()
else:
raise ValueError("Either flow_name or flow_id must be provided")
if not flow:
raise ValueError(f"Flow {flow_name or flow_id} not found")
return self.load_flow(flow.id, tweaks)
def build(self):
raise NotImplementedError

239
src/backend/langflow/interface/custom/directory_reader.py Normal file
View file

@ -0,0 +1,239 @@
import os
import ast
import zlib
class CustomComponentPathValueError(ValueError):
pass
class StringCompressor:
def __init__(self, input_string):
"""Initialize StringCompressor with a string to compress."""
self.input_string = input_string
def compress_string(self):
"""
Compress the initial string and return the compressed data.
"""
# Convert string to bytes
byte_data = self.input_string.encode("utf-8")
# Compress the bytes
self.compressed_data = zlib.compress(byte_data)
return self.compressed_data
def decompress_string(self):
"""
Decompress the compressed data and return the original string.
"""
# Decompress the bytes
decompressed_data = zlib.decompress(self.compressed_data)
# Convert bytes back to string
return decompressed_data.decode("utf-8")
class DirectoryReader:
# Ensure the base path to read the files that contain
# the custom components from this directory.
base_path = ""
def __init__(self, directory_path, compress_code_field=False):
"""
Initialize DirectoryReader with a directory path
and a flag indicating whether to compress the code.
"""
self.directory_path = directory_path
self.compress_code_field = compress_code_field
def get_safe_path(self):
"""Check if the path is valid and return it, or None if it's not."""
return self.directory_path if self.is_valid_path() else None
def is_valid_path(self) -> bool:
"""Check if the directory path is valid by comparing it to the base path."""
fullpath = os.path.normpath(os.path.join(self.directory_path))
return fullpath.startswith(self.base_path)
def is_empty_file(self, file_content):
"""
Check if the file content is empty.
"""
return len(file_content.strip()) == 0
def filter_loaded_components(self, data: dict, with_errors: bool) -> dict:
items = [
{
"name": menu["name"],
"path": menu["path"],
"components": [
component
for component in menu["components"]
if (component["error"] if with_errors else not component["error"])
],
}
for menu in data["menu"]
]
filtred = [menu for menu in items if menu["components"]]
return {"menu": filtred}
def validate_code(self, file_content):
"""
Validate the Python code by trying to parse it with ast.parse.
"""
try:
ast.parse(file_content)
return True
except SyntaxError:
return False
def validate_build(self, file_content):
"""
Check if the file content contains a function named 'build'.
"""
return "def build" in file_content
def read_file_content(self, file_path):
"""
Read and return the content of a file.
"""
if not os.path.isfile(file_path):
return None
with open(file_path, "r") as file:
return file.read()
def get_files(self):
"""
Walk through the directory path and return a list of all .py files.
"""
if not (safe_path := self.get_safe_path()):
raise CustomComponentPathValueError(
f"The path needs to start with '{self.base_path}'."
)
file_list = []
for root, _, files in os.walk(safe_path):
file_list.extend(
os.path.join(root, filename)
for filename in files
if filename.endswith(".py")
)
return file_list
def find_menu(self, response, menu_name):
"""
Find and return a menu by its name in the response.
"""
return next(
(menu for menu in response["menu"] if menu["name"] == menu_name),
None,
)
def _is_type_hint_imported(self, type_hint_name: str, code: str) -> bool:
"""
Check if a specific type hint is imported
from the typing module in the given code.
"""
module = ast.parse(code)
return any(
isinstance(node, ast.ImportFrom)
and node.module == "typing"
and any(alias.name == type_hint_name for alias in node.names)
for node in ast.walk(module)
)
def _is_type_hint_used_in_args(self, type_hint_name: str, code: str) -> bool:
"""
Check if a specific type hint is used in the
function definitions within the given code.
"""
module = ast.parse(code)
for node in ast.walk(module):
if isinstance(node, ast.FunctionDef):
for arg in node.args.args:
if self._is_type_hint_in_arg_annotation(
arg.annotation, type_hint_name
):
return True
return False
def _is_type_hint_in_arg_annotation(self, annotation, type_hint_name: str) -> bool:
"""
Helper function to check if a type hint exists in an annotation.
"""
return (
annotation is not None
and isinstance(annotation, ast.Subscript)
and isinstance(annotation.value, ast.Name)
and annotation.value.id == type_hint_name
)
def is_type_hint_used_but_not_imported(
self, type_hint_name: str, code: str
) -> bool:
"""
Check if a type hint is used but not imported in the given code.
"""
try:
return self._is_type_hint_used_in_args(
type_hint_name, code
) and not self._is_type_hint_imported(type_hint_name, code)
except SyntaxError:
# Returns True if there's something wrong with the code
# TODO : Find a better way to handle this
return True
def process_file(self, file_path):
"""
Process a file by validating its content and
returning the result and content/error message.
"""
file_content = self.read_file_content(file_path)
if file_content is None:
return False, f"Could not read {file_path}"
elif self.is_empty_file(file_content):
return False, "Empty file"
elif not self.validate_code(file_content):
return False, "Syntax error"
elif not self.validate_build(file_content):
return False, "Missing build function"
elif self.is_type_hint_used_but_not_imported("Optional", file_content):
return False, "Type hint 'Optional' is used but not imported in the code."
else:
if self.compress_code_field:
file_content = str(StringCompressor(file_content).compress_string())
return True, file_content
def build_component_menu_list(self, file_paths):
"""
Build a list of menus with their components
from the .py files in the directory.
"""
response = {"menu": []}
for file_path in file_paths:
menu_name = os.path.basename(os.path.dirname(file_path))
filename = os.path.basename(file_path)
validation_result, result_content = self.process_file(file_path)
menu_result = self.find_menu(response, menu_name) or {
"name": menu_name,
"path": os.path.dirname(file_path),
"components": [],
}
component_info = {
"name": filename.split(".")[0],
"file": filename,
"code": result_content if validation_result else "",
"error": "" if validation_result else result_content,
}
menu_result["components"].append(component_info)
if menu_result not in response["menu"]:
response["menu"].append(menu_result)
return response

29
src/backend/langflow/interface/custom/schema.py Normal file
View file

@ -0,0 +1,29 @@
from pydantic import BaseModel, Field
from typing import Optional
class ClassCodeDetails(BaseModel):
"""
A dataclass for storing details about a class.
"""
name: str
doc: Optional[str]
bases: list
attributes: list
methods: list
init: Optional[dict] = Field(default_factory=dict)
class CallableCodeDetails(BaseModel):
"""
A dataclass for storing details about a callable.
"""
name: str
doc: Optional[str]
args: list
body: list
return_type: Optional[str]

14
src/backend/langflow/interface/importing/utils.py
View file

@ -9,6 +9,7 @@ from langchain.base_language import BaseLanguageModel
from langchain.chains.base import Chain
from langchain.chat_models.base import BaseChatModel
from langchain.tools import BaseTool
from langflow.interface.custom.custom_component import CustomComponent
from langflow.utils import validate
from langflow.interface.wrappers.base import wrapper_creator
@ -47,6 +48,7 @@ def import_by_type(_type: str, name: str) -> Any:
"utilities": import_utility,
"output_parsers": import_output_parser,
"retrievers": import_retriever,
"custom_components": import_custom_component,
}
if _type == "llms":
key = "chat" if "chat" in name.lower() else "llm"
@ -57,6 +59,13 @@ def import_by_type(_type: str, name: str) -> Any:
return loaded_func(name)
def import_custom_component(custom_component: str) -> CustomComponent:
"""Import custom component from custom component name"""
return import_class(
f"langflow.interface.custom.custom_component.{custom_component}"
)
def import_output_parser(output_parser: str) -> Any:
"""Import output parser from output parser name"""
return import_module(f"from langchain.output_parsers import {output_parser}")
@ -172,3 +181,8 @@ def get_function(code):
function_name = validate.extract_function_name(code)
return validate.create_function(code, function_name)
def get_function_custom(code):
class_name = validate.extract_class_name(code)
return validate.create_class(code, class_name)

87
src/backend/langflow/interface/initialize/loading.py
View file

@ -1,21 +1,23 @@
import contextlib
import json
from typing import Any, Callable, Dict, List, Sequence, Type
from typing import Any, Callable, Dict, Sequence, Type
from langchain.agents import ZeroShotAgent
from langchain.agents import agent as agent_module
from langchain.agents.agent import AgentExecutor
from langchain.agents.agent_toolkits.base import BaseToolkit
from langchain.agents.tools import BaseTool
from langflow.interface.initialize.llm import initialize_vertexai
from langflow.interface.initialize.utils import handle_format_kwargs, handle_node_type
from langflow.interface.initialize.vector_store import vecstore_initializer
from langchain.schema import Document, BaseOutputParser
from pydantic import ValidationError
from langflow.interface.importing.utils import (
get_function,
get_function_custom,
import_by_type,
)
from langflow.interface.custom_lists import CUSTOM_NODES
from langflow.interface.importing.utils import get_function, import_by_type
from langflow.interface.agents.base import agent_creator
from langflow.interface.toolkits.base import toolkits_creator
from langflow.interface.chains.base import chain_creator
@ -95,12 +97,21 @@ def instantiate_based_on_type(class_object, base_type, node_type, params):
return instantiate_retriever(node_type, class_object, params)
elif base_type == "memory":
return instantiate_memory(node_type, class_object, params)
elif base_type == "custom_components":
return instantiate_custom_component(node_type, class_object, params)
elif base_type == "wrappers":
return instantiate_wrapper(node_type, class_object, params)
else:
return class_object(**params)
def instantiate_custom_component(node_type, class_object, params):
class_object = get_function_custom(params.pop("code"))
custom_component = class_object()
built_object = custom_component.build(**params)
return built_object, {"repr": custom_component.custom_repr()}
def instantiate_wrapper(node_type, class_object, params):
if node_type in wrapper_creator.from_method_nodes:
method = wrapper_creator.from_method_nodes[node_type]
@ -199,68 +210,8 @@ def instantiate_agent(node_type, class_object: Type[agent_module.Agent], params:
def instantiate_prompt(node_type, class_object, params: Dict):
if node_type == "ZeroShotPrompt":
if "tools" not in params:
params["tools"] = []
return ZeroShotAgent.create_prompt(**params)
elif "MessagePromptTemplate" in node_type:
# Then we only need the template
from_template_params = {
"template": params.pop("prompt", params.pop("template", ""))
}
if not from_template_params.get("template"):
raise ValueError("Prompt template is required")
prompt = class_object.from_template(**from_template_params)
elif node_type == "ChatPromptTemplate":
prompt = class_object.from_messages(**params)
else:
prompt = class_object(**params)
format_kwargs: Dict[str, Any] = {}
for input_variable in prompt.input_variables:
if input_variable in params:
variable = params[input_variable]
if isinstance(variable, str):
format_kwargs[input_variable] = variable
elif isinstance(variable, BaseOutputParser) and hasattr(
variable, "get_format_instructions"
):
format_kwargs[input_variable] = variable.get_format_instructions()
elif isinstance(variable, List) and all(
isinstance(item, Document) for item in variable
):
# Format document to contain page_content and metadata
# as one string separated by a newline
if len(variable) > 1:
content = "\n".join(
[item.page_content for item in variable if item.page_content]
)
else:
content = variable[0].page_content
# content could be a json list of strings
with contextlib.suppress(json.JSONDecodeError):
content = json.loads(content)
if isinstance(content, list):
content = ",".join([str(item) for item in content])
format_kwargs[input_variable] = content
# handle_keys will be a list but it does not exist yet
# so we need to create it
if (
isinstance(variable, List)
and all(isinstance(item, Document) for item in variable)
) or (
isinstance(variable, BaseOutputParser)
and hasattr(variable, "get_format_instructions")
):
if "handle_keys" not in format_kwargs:
format_kwargs["handle_keys"] = []
# Add the handle_keys to the list
format_kwargs["handle_keys"].append(input_variable)
params, prompt = handle_node_type(node_type, class_object, params)
format_kwargs = handle_format_kwargs(prompt, params)
return prompt, format_kwargs
@ -363,6 +314,8 @@ def instantiate_textsplitter(
):
try:
documents = params.pop("documents")
if not isinstance(documents, list):
documents = [documents]
except KeyError as exc:
raise ValueError(
"The source you provided did not load correctly or was empty."

103
src/backend/langflow/interface/initialize/utils.py Normal file
View file

@ -0,0 +1,103 @@
import contextlib
import json
from typing import Any, Dict, List
from langchain.agents import ZeroShotAgent
from langchain.schema import Document, BaseOutputParser
def handle_node_type(node_type, class_object, params: Dict):
if node_type == "ZeroShotPrompt":
params = check_tools_in_params(params)
prompt = ZeroShotAgent.create_prompt(**params)
elif "MessagePromptTemplate" in node_type:
prompt = instantiate_from_template(class_object, params)
elif node_type == "ChatPromptTemplate":
prompt = class_object.from_messages(**params)
else:
prompt = class_object(**params)
return params, prompt
def check_tools_in_params(params: Dict):
if "tools" not in params:
params["tools"] = []
return params
def instantiate_from_template(class_object, params: Dict):
from_template_params = {
"template": params.pop("prompt", params.pop("template", ""))
}
if not from_template_params.get("template"):
raise ValueError("Prompt template is required")
return class_object.from_template(**from_template_params)
def handle_format_kwargs(prompt, params: Dict):
format_kwargs: Dict[str, Any] = {}
for input_variable in prompt.input_variables:
if input_variable in params:
format_kwargs = handle_variable(params, input_variable, format_kwargs)
return format_kwargs
def handle_variable(params: Dict, input_variable: str, format_kwargs: Dict):
variable = params[input_variable]
if isinstance(variable, str):
format_kwargs[input_variable] = variable
elif isinstance(variable, BaseOutputParser) and hasattr(
variable, "get_format_instructions"
):
format_kwargs[input_variable] = variable.get_format_instructions()
elif is_instance_of_list_or_document(variable):
format_kwargs = format_document(variable, input_variable, format_kwargs)
if needs_handle_keys(variable):
format_kwargs = add_handle_keys(input_variable, format_kwargs)
return format_kwargs
def is_instance_of_list_or_document(variable):
return (
isinstance(variable, List)
and all(isinstance(item, Document) for item in variable)
or isinstance(variable, Document)
)
def format_document(variable, input_variable: str, format_kwargs: Dict):
variable = variable if isinstance(variable, List) else [variable]
content = format_content(variable)
format_kwargs[input_variable] = content
return format_kwargs
def format_content(variable):
if len(variable) > 1:
return "\n".join([item.page_content for item in variable if item.page_content])
content = variable[0].page_content
return try_to_load_json(content)
def try_to_load_json(content):
with contextlib.suppress(json.JSONDecodeError):
content = json.loads(content)
if isinstance(content, list):
content = ",".join([str(item) for item in content])
return content
def needs_handle_keys(variable):
return is_instance_of_list_or_document(variable) or (
isinstance(variable, BaseOutputParser)
and hasattr(variable, "get_format_instructions")
)
def add_handle_keys(input_variable: str, format_kwargs: Dict):
if "handle_keys" not in format_kwargs:
format_kwargs["handle_keys"] = []
format_kwargs["handle_keys"].append(input_variable)
return format_kwargs

2
src/backend/langflow/interface/listing.py
View file

@ -13,6 +13,7 @@ from langflow.interface.vector_store.base import vectorstore_creator
from langflow.interface.wrappers.base import wrapper_creator
from langflow.interface.output_parsers.base import output_parser_creator
from langflow.interface.retrievers.base import retriever_creator
from langflow.interface.custom.base import custom_component_creator
def get_type_dict():
@ -32,6 +33,7 @@ def get_type_dict():
"utilities": utility_creator.to_list(),
"outputParsers": output_parser_creator.to_list(),
"retrievers": retriever_creator.to_list(),
"custom_components": custom_component_creator.to_list(),
}

2
src/backend/langflow/interface/tools/base.py
View file

@ -55,7 +55,7 @@ TOOL_INPUTS = {
show=True,
value="",
suffixes=[".json", ".yaml", ".yml"],
fileTypes=["json", "yaml", "yml"],
file_types=["json", "yaml", "yml"],
),
}

5
src/backend/langflow/interface/tools/constants.py
View file

@ -9,7 +9,10 @@ from langchain.agents.load_tools import (
from langchain.tools.json.tool import JsonSpec
from langflow.interface.importing.utils import import_class
from langflow.interface.tools.custom import PythonFunctionTool, PythonFunction
from langflow.interface.tools.custom import (
PythonFunctionTool,
PythonFunction,
)
FILE_TOOLS = {"JsonSpec": JsonSpec}
CUSTOM_TOOLS = {

4
src/backend/langflow/interface/tools/custom.py
View file

@ -34,8 +34,6 @@ class Function(BaseModel):
class PythonFunctionTool(Function, Tool):
"""Python function"""
name: str = "Custom Tool"
description: str
code: str
@ -49,6 +47,4 @@ class PythonFunctionTool(Function, Tool):
class PythonFunction(Function):
"""Python function"""
code: str

334
src/backend/langflow/interface/types.py
View file

@ -1,7 +1,10 @@
from typing import Any
from langflow.interface.agents.base import agent_creator
from langflow.interface.chains.base import chain_creator
from langflow.interface.custom.constants import CUSTOM_COMPONENT_SUPPORTED_TYPES
from langflow.interface.document_loaders.base import documentloader_creator
from langflow.interface.embeddings.base import embedding_creator
from langflow.interface.importing.utils import get_function_custom
from langflow.interface.llms.base import llm_creator
from langflow.interface.memories.base import memory_creator
from langflow.interface.prompts.base import prompt_creator
@ -12,9 +15,28 @@ from langflow.interface.utilities.base import utility_creator
from langflow.interface.vector_store.base import vectorstore_creator
from langflow.interface.wrappers.base import wrapper_creator
from langflow.interface.output_parsers.base import output_parser_creator
from langflow.interface.custom.base import custom_component_creator
from langflow.interface.custom.custom_component import CustomComponent
from langflow.template.field.base import TemplateField
from langflow.template.frontend_node.constants import CLASSES_TO_REMOVE
from langflow.template.frontend_node.custom_components import (
CustomComponentFrontendNode,
)
from langflow.interface.retrievers.base import retriever_creator
from langflow.interface.custom.directory_reader import DirectoryReader
from langflow.utils.logger import logger
from langflow.utils.util import get_base_classes
from langflow.api.utils import merge_nested_dicts
import re
import warnings
import traceback
from fastapi import HTTPException
# Used to get the base_classes list
def get_type_list():
"""Get a list of all langchain types"""
all_types = build_langchain_types_dict()
@ -29,7 +51,6 @@ def get_type_list():
def build_langchain_types_dict(): # sourcery skip: dict-assign-update-to-union
"""Build a dictionary of all langchain types"""
all_types = {}
creators = [
@ -48,6 +69,7 @@ def build_langchain_types_dict(): # sourcery skip: dict-assign-update-to-union
utility_creator,
output_parser_creator,
retriever_creator,
custom_component_creator,
]
all_types = {}
@ -55,7 +77,315 @@ def build_langchain_types_dict(): # sourcery skip: dict-assign-update-to-union
created_types = creator.to_dict()
if created_types[creator.type_name].values():
all_types.update(created_types)
return all_types
langchain_types_dict = build_langchain_types_dict()
def process_type(field_type: str):
return "prompt" if field_type == "Prompt" else field_type
# TODO: Move to correct place
def add_new_custom_field(
template,
field_name: str,
field_type: str,
field_value: Any,
field_required: bool,
field_config: dict,
):
# Check field_config if any of the keys are in it
# if it is, update the value
display_name = field_config.pop("display_name", field_name)
field_type = field_config.pop("field_type", field_type)
field_type = process_type(field_type)
field_value = field_config.pop("value", field_value)
field_advanced = field_config.pop("advanced", False)
if field_type == "bool" and field_value is None:
field_value = False
# If options is a list, then it's a dropdown
# If options is None, then it's a list of strings
is_list = isinstance(field_config.get("options"), list)
field_config["is_list"] = is_list or field_config.get("is_list", False)
if "name" in field_config:
warnings.warn(
"The 'name' key in field_config is used to build the object and can't be changed."
)
field_config.pop("name", None)
required = field_config.pop("required", field_required)
placeholder = field_config.pop("placeholder", "")
new_field = TemplateField(
name=field_name,
field_type=field_type,
value=field_value,
show=True,
required=required,
advanced=field_advanced,
placeholder=placeholder,
display_name=display_name,
**field_config,
)
template.get("template")[field_name] = new_field.to_dict()
template.get("custom_fields")[field_name] = None
return template
# TODO: Move to correct place
def add_code_field(template, raw_code, field_config):
# Field with the Python code to allow update
code_field = {
"code": {
"dynamic": True,
"required": True,
"placeholder": "",
"show": True,
"multiline": True,
"value": raw_code,
"password": False,
"name": "code",
"advanced": field_config.pop("advanced", False),
"type": "code",
"list": False,
}
}
template.get("template")["code"] = code_field.get("code")
return template
def extract_type_from_optional(field_type):
"""
Extract the type from a string formatted as "Optional[<type>]".
Parameters:
field_type (str): The string from which to extract the type.
Returns:
str: The extracted type, or an empty string if no type was found.
"""
match = re.search(r"\[(.*?)\]", field_type)
return match[1] if match else None
def build_frontend_node(custom_component: CustomComponent):
"""Build a frontend node for a custom component"""
try:
return (
CustomComponentFrontendNode().to_dict().get(type(custom_component).__name__)
)
except Exception as exc:
logger.error(f"Error while building base frontend node: {exc}")
return None
def update_display_name_and_description(frontend_node, template_config):
"""Update the display name and description of a frontend node"""
if "display_name" in template_config:
frontend_node["display_name"] = template_config["display_name"]
if "description" in template_config:
frontend_node["description"] = template_config["description"]
def build_field_config(custom_component: CustomComponent):
"""Build the field configuration for a custom component"""
try:
custom_class = get_function_custom(custom_component.code)
except Exception as exc:
logger.error(f"Error while getting custom function: {str(exc)}")
return {}
try:
return custom_class().build_config()
except Exception as exc:
logger.error(f"Error while building field config: {str(exc)}")
return {}
def add_extra_fields(frontend_node, field_config, function_args):
"""Add extra fields to the frontend node"""
if function_args is None or function_args == "":
return
# sort function_args which is a list of dicts
function_args.sort(key=lambda x: x["name"])
for extra_field in function_args:
if "name" not in extra_field or extra_field["name"] == "self":
continue
field_name, field_type, field_value, field_required = get_field_properties(
extra_field
)
config = field_config.get(field_name, {})
frontend_node = add_new_custom_field(
frontend_node,
field_name,
field_type,
field_value,
field_required,
config,
)
def get_field_properties(extra_field):
"""Get the properties of an extra field"""
field_name = extra_field["name"]
field_type = extra_field.get("type", "str")
field_value = extra_field.get("default", "")
field_required = "optional" not in field_type.lower()
if not field_required:
field_type = extract_type_from_optional(field_type)
return field_name, field_type, field_value, field_required
def add_base_classes(frontend_node, return_type):
"""Add base classes to the frontend node"""
if return_type not in CUSTOM_COMPONENT_SUPPORTED_TYPES or return_type is None:
raise HTTPException(
status_code=400,
detail={
"error": (
"Invalid return type should be one of: "
f"{list(CUSTOM_COMPONENT_SUPPORTED_TYPES.keys())}"
),
"traceback": traceback.format_exc(),
},
)
return_type_instance = CUSTOM_COMPONENT_SUPPORTED_TYPES.get(return_type)
base_classes = get_base_classes(return_type_instance)
for base_class in base_classes:
if base_class not in CLASSES_TO_REMOVE:
frontend_node.get("base_classes").append(base_class)
def build_langchain_template_custom_component(custom_component: CustomComponent):
"""Build a custom component template for the langchain"""
frontend_node = build_frontend_node(custom_component)
if frontend_node is None:
return None
template_config = custom_component.build_template_config
update_display_name_and_description(frontend_node, template_config)
field_config = build_field_config(custom_component)
add_extra_fields(
frontend_node, field_config, custom_component.get_function_entrypoint_args
)
frontend_node = add_code_field(
frontend_node, custom_component.code, field_config.get("code", {})
)
add_base_classes(
frontend_node, custom_component.get_function_entrypoint_return_type
)
return frontend_node
def load_files_from_path(path: str):
"""Load all files from a given path"""
reader = DirectoryReader(path, False)
return reader.get_files()
def build_and_validate_all_files(reader, file_list):
"""Build and validate all files"""
data = reader.build_component_menu_list(file_list)
valid_components = reader.filter_loaded_components(data=data, with_errors=False)
invalid_components = reader.filter_loaded_components(data=data, with_errors=True)
return valid_components, invalid_components
def build_valid_menu(valid_components):
"""Build the valid menu"""
valid_menu = {}
for menu_item in valid_components["menu"]:
menu_name = menu_item["name"]
valid_menu[menu_name] = {}
for component in menu_item["components"]:
try:
component_name = component["name"]
component_code = component["code"]
component_extractor = CustomComponent(code=component_code)
component_extractor.is_check_valid()
component_template = build_langchain_template_custom_component(
component_extractor
)
valid_menu[menu_name][component_name] = component_template
except Exception as exc:
logger.error(f"Error while building custom component: {exc}")
return valid_menu
def build_invalid_menu(invalid_components):
"""Build the invalid menu"""
invalid_menu = {}
for menu_item in invalid_components["menu"]:
menu_name = menu_item["name"]
invalid_menu[menu_name] = {}
for component in menu_item["components"]:
try:
component_name = component["name"]
component_code = component["code"]
component_template = (
CustomComponentFrontendNode(
description="ERROR - Check your Python Code",
display_name=f"ERROR - {component_name}",
)
.to_dict()
.get(type(CustomComponent()).__name__)
)
component_template["error"] = component.get("error", None)
component_template.get("template").get("code")["value"] = component_code
invalid_menu[menu_name][component_name] = component_template
except Exception as exc:
logger.error(
f"Error while creating custom component [{component_name}]: {str(exc)}"
)
return invalid_menu
def build_langchain_custom_component_list_from_path(path: str):
"""Build a list of custom components for the langchain from a given path"""
file_list = load_files_from_path(path)
reader = DirectoryReader(path, False)
valid_components, invalid_components = build_and_validate_all_files(
reader, file_list
)
valid_menu = build_valid_menu(valid_components)
invalid_menu = build_invalid_menu(invalid_components)
return merge_nested_dicts(valid_menu, invalid_menu)

16
src/backend/langflow/main.py
View file

@ -8,10 +8,12 @@ from fastapi.staticfiles import StaticFiles
from langflow.api import router
from langflow.database.base import create_db_and_tables
from langflow.interface.utils import setup_llm_caching
from langflow.utils.logger import configure
def create_app():
"""Create the FastAPI app and include the router."""
configure()
app = FastAPI()
@ -78,10 +80,16 @@ def setup_app(static_files_dir: Optional[Path] = None) -> FastAPI:
return app
app = create_app()
if __name__ == "__main__":
import uvicorn
from langflow.utils.util import get_number_of_workers
uvicorn.run(app, host="127.0.0.1", port=7860)
configure()
uvicorn.run(
create_app,
host="127.0.0.1",
port=7860,
workers=get_number_of_workers(),
log_level="debug",
reload=True,
)

2
src/backend/langflow/processing/base.py
View file

@ -22,7 +22,7 @@ async def get_result_and_steps(langchain_object, inputs: Union[dict, str], **kwa
try:
fix_memory_inputs(langchain_object)
except Exception as exc:
logger.error(exc)
logger.error(f"Error fixing memory inputs: {exc}")
try:
async_callbacks = [AsyncStreamingLLMCallbackHandler(**kwargs)]

7
src/backend/langflow/processing/process.py
View file

@ -85,12 +85,17 @@ def get_input_str_if_only_one_input(inputs: dict) -> Optional[str]:
return list(inputs.values())[0] if len(inputs) == 1 else None
def process_graph_cached(data_graph: Dict[str, Any], inputs: Optional[dict] = None):
def process_graph_cached(
data_graph: Dict[str, Any], inputs: Optional[dict] = None, clear_cache=False
):
"""
Process graph by extracting input variables and replacing ZeroShotPrompt
with PromptTemplate,then run the graph and return the result and thought.
"""
# Load langchain object
if clear_cache:
build_sorted_vertices_with_caching.clear_cache()
logger.debug("Cleared cache")
langchain_object, artifacts = build_sorted_vertices_with_caching(data_graph)
logger.debug("Loaded LangChain object")
if inputs is None:

37
src/backend/langflow/settings.py
View file

@ -1,10 +1,13 @@
import os
from typing import Optional
from typing import Optional, List
from pathlib import Path
import yaml
from pydantic import BaseSettings, root_validator
from langflow.utils.logger import logger
BASE_COMPONENTS_PATH = Path(__file__).parent / "components"
class Settings(BaseSettings):
chains: dict = {}
@ -22,13 +25,16 @@ class Settings(BaseSettings):
textsplitters: dict = {}
utilities: dict = {}
output_parsers: dict = {}
custom_components: dict = {}
dev: bool = False
database_url: Optional[str] = None
cache: str = "InMemoryCache"
remove_api_keys: bool = False
components_path: List[Path]
@root_validator(pre=True)
def set_database_url(cls, values):
def set_env_variables(cls, values):
if "database_url" not in values:
logger.debug(
"No database_url provided, trying LANGFLOW_DATABASE_URL env variable"
@ -38,6 +44,23 @@ class Settings(BaseSettings):
else:
logger.debug("No DATABASE_URL env variable, using sqlite database")
values["database_url"] = "sqlite:///./langflow.db"
if not values.get("components_path"):
values["components_path"] = [BASE_COMPONENTS_PATH]
logger.debug("No components_path provided, using default components path")
elif BASE_COMPONENTS_PATH not in values["components_path"]:
values["components_path"].append(BASE_COMPONENTS_PATH)
logger.debug("Adding default components path to components_path")
if os.getenv("LANGFLOW_COMPONENTS_PATH"):
logger.debug("Adding LANGFLOW_COMPONENTS_PATH to components_path")
langflow_component_path = Path(os.getenv("LANGFLOW_COMPONENTS_PATH"))
if (
langflow_component_path.exists()
and langflow_component_path not in values["components_path"]
):
values["components_path"].append(langflow_component_path)
logger.debug(f"Adding {langflow_component_path} to components_path")
return values
class Config:
@ -68,12 +91,20 @@ class Settings(BaseSettings):
self.documentloaders = new_settings.documentloaders or {}
self.retrievers = new_settings.retrievers or {}
self.output_parsers = new_settings.output_parsers or {}
self.custom_components = new_settings.custom_components or {}
self.components_path = new_settings.components_path or []
self.dev = dev
def update_settings(self, **kwargs):
for key, value in kwargs.items():
if hasattr(self, key):
setattr(self, key, value)
if isinstance(getattr(self, key), list):
if isinstance(value, list):
getattr(self, key).extend(value)
else:
getattr(self, key).append(value)
else:
setattr(self, key, value)
def save_settings_to_yaml(settings: Settings, file_path: str):

37
src/backend/langflow/template/field/base.py
View file

@ -6,23 +6,58 @@ from pydantic import BaseModel
class TemplateFieldCreator(BaseModel, ABC):
field_type: str = "str"
"""The type of field this is. Default is a string."""
required: bool = False
"""Specifies if the field is required. Defaults to False."""
placeholder: str = ""
"""A placeholder string for the field. Default is an empty string."""
is_list: bool = False
"""Defines if the field is a list. Default is False."""
show: bool = True
"""Should the field be shown. Defaults to True."""
multiline: bool = False
"""Defines if the field will allow the user to open a text editor. Default is False."""
value: Any = None
"""The value of the field. Default is None."""
suffixes: list[str] = []
fileTypes: list[str] = []
"""List of suffixes for a file field. Default is an empty list."""
file_types: list[str] = []
"""List of file types associated with the field. Default is an empty list. (duplicate)"""
file_path: Union[str, None] = None
"""The file path of the field if it is a file. Defaults to None."""
password: bool = False
"""Specifies if the field is a password. Defaults to False."""
options: list[str] = []
"""List of options for the field. Only used when is_list=True. Default is an empty list."""
name: str = ""
"""Name of the field. Default is an empty string."""
display_name: Optional[str] = None
"""Display name of the field. Defaults to None."""
advanced: bool = False
"""Specifies if the field will an advanced parameter (hidden). Defaults to False."""
input_types: list[str] = []
"""List of input types for the handle when the field has more than one type. Default is an empty list."""
dynamic: bool = False
"""Specifies if the field is dynamic. Defaults to False."""
info: Optional[str] = ""
"""Additional information about the field to be shown in the tooltip. Defaults to an empty string."""
def to_dict(self):
result = self.dict()

2
src/backend/langflow/template/frontend_node/__init__.py
View file

@ -9,6 +9,7 @@ from langflow.template.frontend_node import (
vectorstores,
documentloaders,
textsplitters,
custom_components,
)
__all__ = [
@ -22,4 +23,5 @@ __all__ = [
"vectorstores",
"documentloaders",
"textsplitters",
"custom_components",
]

2
src/backend/langflow/template/frontend_node/agents.py
View file

@ -145,7 +145,7 @@ class CSVAgentNode(FrontendNode):
name="path",
value="",
suffixes=[".csv"],
fileTypes=["csv"],
file_types=["csv"],
),
TemplateField(
field_type="BaseLanguageModel",

27
src/backend/langflow/template/frontend_node/base.py
View file

@ -5,13 +5,14 @@ from typing import List, Optional
from pydantic import BaseModel, Field
from langflow.template.frontend_node.formatter import field_formatters
from langflow.template.frontend_node.constants import FORCE_SHOW_FIELDS
from langflow.template.frontend_node.constants import (
CLASSES_TO_REMOVE,
FORCE_SHOW_FIELDS,
)
from langflow.template.field.base import TemplateField
from langflow.template.template.base import Template
from langflow.utils import constants
CLASSES_TO_REMOVE = ["Serializable", "BaseModel", "object"]
class FieldFormatters(BaseModel):
formatters = {
@ -51,14 +52,8 @@ class FrontendNode(BaseModel):
custom_fields: defaultdict = defaultdict(list)
output_types: List[str] = []
field_formatters: FieldFormatters = Field(default_factory=FieldFormatters)
def process_base_classes(self) -> None:
"""Removes unwanted base classes from the list of base classes."""
self.base_classes = [
base_class
for base_class in self.base_classes
if base_class not in CLASSES_TO_REMOVE
]
beta: bool = False
error: Optional[str] = None
# field formatters is an instance attribute but it is not used in the class
# so we need to create a method to get it
@ -70,6 +65,14 @@ class FrontendNode(BaseModel):
"""Sets the documentation of the frontend node."""
self.documentation = documentation
def process_base_classes(self) -> None:
"""Removes unwanted base classes from the list of base classes."""
self.base_classes = [
base_class
for base_class in self.base_classes
if base_class not in CLASSES_TO_REMOVE
]
def to_dict(self) -> dict:
"""Returns a dict representation of the frontend node."""
self.process_base_classes()
@ -82,6 +85,8 @@ class FrontendNode(BaseModel):
"custom_fields": self.custom_fields,
"output_types": self.output_types,
"documentation": self.documentation,
"beta": self.beta,
"error": self.error,
},
}

3
src/backend/langflow/template/frontend_node/constants.py
View file

@ -63,3 +63,6 @@ You can change this to use other APIs like JinaChat, LocalAI and Prem.
INPUT_KEY_INFO = """The variable to be used as Chat Input when more than one variable is available."""
OUTPUT_KEY_INFO = """The variable to be used as Chat Output (e.g. answer in a ConversationalRetrievalChain)"""
CLASSES_TO_REMOVE = ["Serializable", "BaseModel", "object"]

31
src/backend/langflow/template/frontend_node/custom_components.py Normal file
View file

@ -0,0 +1,31 @@
from langflow.template.field.base import TemplateField
from langflow.template.frontend_node.base import FrontendNode
from langflow.template.template.base import Template
from langflow.interface.custom.constants import DEFAULT_CUSTOM_COMPONENT_CODE
class CustomComponentFrontendNode(FrontendNode):
name: str = "CustomComponent"
display_name: str = "Custom Component"
beta: bool = True
template: Template = Template(
type_name="CustomComponent",
fields=[
TemplateField(
field_type="code",
required=True,
placeholder="",
is_list=False,
show=True,
value=DEFAULT_CUSTOM_COMPONENT_CODE,
name="code",
advanced=False,
dynamic=True,
)
],
)
description: str = "Create any custom component you want!"
base_classes: list[str] = []
def to_dict(self):
return super().to_dict()

2
src/backend/langflow/template/frontend_node/documentloaders.py
View file

@ -14,7 +14,7 @@ def build_file_field(
name=name,
value="",
suffixes=suffixes,
fileTypes=fileTypes,
file_types=fileTypes,
)

Some files were not shown because too many files have changed in this diff Show more