feat: add starter project graphs (#3369)

* feat: Add basic prompting graph function.

* feat: Add blog writer starter project function.

* feat(langflow): Add document QA starter project.

* feat: Add memory chatbot graph function to create chatbot with memory component.

* feat: Add hierarchical tasks agent graph to handle sequential tasks.

* feat: Add a function to create a sequential tasks agent with specific tasks.

* feat: Add vector_store_rag module with ingestion and RAG graphs.

* Refactor: Update the hierarchical task agent to use builder methods for agents and models.

* feat: Refactor sequential tasks agent to utilize build_model and build_output methods.

* refactor: Rename functions in blog_writer, document_qa, and vector_store_rag to end with "_graph".

* feat: Add new graphs to starter projects __init__.py.

* feat: Add complex agent graph setup with prompts, tools, and agents.

* refactor: Add complex agent graph to starter projects.

* feat: Add starter project graphs and dump retrieval functions.

* test: Refactor test_directory_without_mocks method with temporary directory for testing purposes.

src/backend/base/langflow/initial_setup/load.py

@@ -0,0 +1,25 @@
+from .starter_projects import (
+    basic_prompting_graph,
+    blog_writer_graph,
+    document_qa_graph,
+    hierarchical_tasks_agent_graph,
+    memory_chatbot_graph,
+    sequential_tasks_agent_graph,
+    vector_store_rag_graph,
+)
+
+
+def get_starter_projects_graphs():
+    return [
+        basic_prompting_graph(),
+        blog_writer_graph(),
+        document_qa_graph(),
+        memory_chatbot_graph(),
+        vector_store_rag_graph(),
+        sequential_tasks_agent_graph(),
+        hierarchical_tasks_agent_graph(),
+    ]
+
+
+def get_starter_projects_dump():
+    return [g.dump() for g in get_starter_projects_graphs()]

src/backend/base/langflow/initial_setup/starter_projects/__init__.py

@@ -0,0 +1,19 @@
+from .basic_prompting import basic_prompting_graph
+from .blog_writer import blog_writer_graph
+from .document_qa import document_qa_graph
+from .hierarchical_tasks_agent import hierarchical_tasks_agent_graph
+from .memory_chatbot import memory_chatbot_graph
+from .sequential_tasks_agent import sequential_tasks_agent_graph
+from .vector_store_rag import vector_store_rag_graph
+from .complex_agent import complex_agent_graph
+
+__all__ = [
+    "blog_writer_graph",
+    "document_qa_graph",
+    "memory_chatbot_graph",
+    "vector_store_rag_graph",
+    "basic_prompting_graph",
+    "sequential_tasks_agent_graph",
+    "hierarchical_tasks_agent_graph",
+    "complex_agent_graph",
+]

src/backend/base/langflow/initial_setup/starter_projects/basic_prompting.py

@@ -0,0 +1,30 @@
+from langflow.components.inputs.ChatInput import ChatInput
+from langflow.components.models.OpenAIModel import OpenAIModelComponent
+from langflow.components.outputs.ChatOutput import ChatOutput
+from langflow.components.prompts.Prompt import PromptComponent
+from langflow.graph.graph.base import Graph
+
+
+def basic_prompting_graph(template: str | None = None):
+    if template is None:
+        template = """Answer the user as if you were a pirate.
+
+User: {user_input}
+
+Answer:
+"""
+    chat_input = ChatInput()
+    prompt_component = PromptComponent()
+    prompt_component.set(
+        template=template,
+        user_input=chat_input.message_response,
+    )
+
+    openai_component = OpenAIModelComponent()
+    openai_component.set(input_value=prompt_component.build_prompt)
+
+    chat_output = ChatOutput()
+    chat_output.set(input_value=openai_component.text_response)
+
+    graph = Graph(start=chat_input, end=chat_output)
+    return graph

src/backend/base/langflow/initial_setup/starter_projects/blog_writer.py

@@ -0,0 +1,48 @@
+from textwrap import dedent
+
+from langflow.components.data.URL import URLComponent
+from langflow.components.helpers.ParseData import ParseDataComponent
+from langflow.components.inputs.TextInput import TextInputComponent
+from langflow.components.models.OpenAIModel import OpenAIModelComponent
+from langflow.components.outputs.ChatOutput import ChatOutput
+from langflow.components.prompts.Prompt import PromptComponent
+from langflow.graph.graph.base import Graph
+
+
+def blog_writer_graph(template: str | None = None):
+    if template is None:
+        template = dedent("""Reference 1:
+
+{references}
+
+---
+
+{instructions}
+
+Blog:
+""")
+    url_component = URLComponent()
+    url_component.set(urls=["https://langflow.org/", "https://docs.langflow.org/"])
+    parse_data_component = ParseDataComponent()
+    parse_data_component.set(data=url_component.fetch_content)
+
+    text_input = TextInputComponent(_display_name="Instructions")
+    text_input.set(
+        input_value="Use the references above for style to write a new blog/tutorial about Langflow and AI. Suggest non-covered topics."
+    )
+
+    prompt_component = PromptComponent()
+    prompt_component.set(
+        template=template,
+        instructions=text_input.text_response,
+        references=parse_data_component.parse_data,
+    )
+
+    openai_component = OpenAIModelComponent()
+    openai_component.set(input_value=prompt_component.build_prompt)
+
+    chat_output = ChatOutput()
+    chat_output.set(input_value=openai_component.text_response)
+
+    graph = Graph(start=text_input, end=chat_output)
+    return graph

src/backend/base/langflow/initial_setup/starter_projects/complex_agent.py

@@ -0,0 +1,98 @@
+from langflow.components.agents.CrewAIAgent import CrewAIAgentComponent
+from langflow.components.agents.HierarchicalCrew import HierarchicalCrewComponent
+from langflow.components.helpers.HierarchicalTask import HierarchicalTaskComponent
+from langflow.components.inputs.ChatInput import ChatInput
+from langflow.components.models.OpenAIModel import OpenAIModelComponent
+from langflow.components.outputs.ChatOutput import ChatOutput
+from langflow.components.prompts.Prompt import PromptComponent
+from langflow.components.tools.SearchAPI import SearchAPIComponent
+from langflow.components.tools.YfinanceTool import YfinanceToolComponent
+from langflow.graph.graph.base import Graph
+
+
+def complex_agent_graph():
+    llm = OpenAIModelComponent(model_name="gpt-4o-mini")
+    manager_llm = OpenAIModelComponent(model_name="gpt-4o")
+    search_api_tool = SearchAPIComponent()
+    yahoo_search_tool = YfinanceToolComponent()
+    dynamic_agent = CrewAIAgentComponent()
+    chat_input = ChatInput()
+    role_prompt = PromptComponent(_display_name="Role Prompt")
+    role_prompt.set(
+        template="""Define a Role that could execute or answer well the user's query.
+
+User's query: {query}
+
+Role should be two words max. Something like "Researcher" or "Software Developer".
+"""
+    )
+
+    goal_prompt = PromptComponent(_display_name="Goal Prompt")
+    goal_prompt.set(
+        template="""Define the Goal of this Role, given the User's Query.
+User's query: {query}
+
+Role: {role}
+
+The goal should be concise and specific.
+Goal:
+""",
+        query=chat_input.message_response,
+        role=role_prompt.build_prompt,
+    )
+    backstory_prompt = PromptComponent(_display_name="Backstory Prompt")
+    backstory_prompt.set(
+        template="""Define a Backstory of this Role and Goal, given the User's Query.
+User's query: {query}
+
+Role: {role}
+Goal: {goal}
+
+The backstory should be specific and well aligned with the rest of the information.
+Backstory:""",
+        query=chat_input.message_response,
+        role=role_prompt.build_prompt,
+        goal=goal_prompt.build_prompt,
+    )
+    dynamic_agent.set(
+        tools=[search_api_tool.build_tool, yahoo_search_tool.build_tool],
+        llm=llm.build_model,
+        role=role_prompt.build_prompt,
+        goal=goal_prompt.build_prompt,
+        backstory=backstory_prompt.build_prompt,
+    )
+
+    response_prompt = PromptComponent()
+    response_prompt.set(
+        template="""User's query:
+{query}
+
+Respond to the user with as much as information as you can about the topic. Delete if needed. If it is just a general query (e.g a greeting) you can respond them directly.""",
+        query=chat_input.message_response,
+    )
+    manager_agent = CrewAIAgentComponent()
+    manager_agent.set(
+        llm=manager_llm.build_model,
+        role="Manager",
+        goal="You can answer general questions from the User and may call others for help if needed.",
+        backstory="You are polite and helpful. You've always been a beacon of politeness.",
+    )
+    task = HierarchicalTaskComponent()
+    task.set(
+        task_description=response_prompt.build_prompt,
+        expected_output="Succinct response that answers the User's query.",
+    )
+    crew_component = HierarchicalCrewComponent()
+    crew_component.set(
+        tasks=task.build_task, agents=[dynamic_agent.build_output], manager_agent=manager_agent.build_output
+    )
+    chat_output = ChatOutput()
+    chat_output.set(input_value=crew_component.build_output)
+
+    graph = Graph(
+        start=chat_input,
+        end=chat_output,
+        flow_name="Sequential Tasks Agent",
+        description="This Agent runs tasks in a predefined sequence.",
+    )
+    return graph

src/backend/base/langflow/initial_setup/starter_projects/document_qa.py

@@ -0,0 +1,44 @@
+from langflow.components.data.File import FileComponent
+from langflow.components.helpers.ParseData import ParseDataComponent
+from langflow.components.inputs.ChatInput import ChatInput
+from langflow.components.models.OpenAIModel import OpenAIModelComponent
+from langflow.components.outputs.ChatOutput import ChatOutput
+from langflow.components.prompts.Prompt import PromptComponent
+from langflow.graph.graph.base import Graph
+
+
+def document_qa_graph(template: str | None = None):
+    if template is None:
+        template = """Answer user's questions based on the document below:
+
+---
+
+{Document}
+
+---
+
+Question:
+{Question}
+
+Answer:
+"""
+    file_component = FileComponent()
+    parse_data_component = ParseDataComponent()
+    parse_data_component.set(data=file_component.load_file)
+
+    chat_input = ChatInput()
+    prompt_component = PromptComponent()
+    prompt_component.set(
+        template=template,
+        context=parse_data_component.parse_data,
+        question=chat_input.message_response,
+    )
+
+    openai_component = OpenAIModelComponent()
+    openai_component.set(input_value=prompt_component.build_prompt)
+
+    chat_output = ChatOutput()
+    chat_output.set(input_value=openai_component.text_response)
+
+    graph = Graph(start=chat_input, end=chat_output)
+    return graph

src/backend/base/langflow/initial_setup/starter_projects/hierarchical_tasks_agent.py

@@ -0,0 +1,70 @@
+from langflow.components.agents.CrewAIAgent import CrewAIAgentComponent
+from langflow.components.agents.HierarchicalCrew import HierarchicalCrewComponent
+from langflow.components.helpers.HierarchicalTask import HierarchicalTaskComponent
+from langflow.components.inputs.ChatInput import ChatInput
+from langflow.components.models.OpenAIModel import OpenAIModelComponent
+from langflow.components.outputs.ChatOutput import ChatOutput
+from langflow.components.prompts.Prompt import PromptComponent
+from langflow.components.tools.SearchAPI import SearchAPIComponent
+from langflow.graph.graph.base import Graph
+
+
+def hierarchical_tasks_agent_graph():
+    llm = OpenAIModelComponent(model_name="gpt-4o-mini")
+    manager_llm = OpenAIModelComponent(model_name="gpt-4o")
+    search_api_tool = SearchAPIComponent()
+    researcher_agent = CrewAIAgentComponent()
+    chat_input = ChatInput()
+    researcher_agent.set(
+        tools=[search_api_tool.build_tool],
+        llm=llm.build_model,
+        role="Researcher",
+        goal="Search for information about the User's query and answer as best as you can",
+        backstory="You are a reliable researcher and journalist ",
+    )
+
+    editor_agent = CrewAIAgentComponent()
+
+    editor_agent.set(
+        llm=llm.build_model,
+        role="Editor",
+        goal="Evaluate the information for misleading or biased data.",
+        backstory="You are a reliable researcher and journalist ",
+    )
+
+    response_prompt = PromptComponent()
+    response_prompt.set(
+        template="""User's query:
+{query}
+
+Respond to the user with as much as information as you can about the topic. Delete if needed. If it is just a general query (e.g a greeting) you can respond them directly.""",
+        query=chat_input.message_response,
+    )
+    manager_agent = CrewAIAgentComponent()
+    manager_agent.set(
+        llm=manager_llm.build_model,
+        role="Manager",
+        goal="You can answer general questions from the User and may call others for help if needed.",
+        backstory="You are polite and helpful. You've always been a beacon of politeness.",
+    )
+    task = HierarchicalTaskComponent()
+    task.set(
+        task_description=response_prompt.build_prompt,
+        expected_output="Succinct response that answers the User's query.",
+    )
+    crew_component = HierarchicalCrewComponent()
+    crew_component.set(
+        tasks=task.build_task,
+        agents=[researcher_agent.build_output, editor_agent.build_output],
+        manager_agent=manager_agent.build_output,
+    )
+    chat_output = ChatOutput()
+    chat_output.set(input_value=crew_component.build_output)
+
+    graph = Graph(
+        start=chat_input,
+        end=chat_output,
+        flow_name="Sequential Tasks Agent",
+        description="This Agent runs tasks in a predefined sequence.",
+    )
+    return graph

src/backend/base/langflow/initial_setup/starter_projects/memory_chatbot.py

@@ -0,0 +1,28 @@
+from langflow.components.helpers.Memory import MemoryComponent
+from langflow.components.inputs.ChatInput import ChatInput
+from langflow.components.models.OpenAIModel import OpenAIModelComponent
+from langflow.components.outputs.ChatOutput import ChatOutput
+from langflow.components.prompts.Prompt import PromptComponent
+from langflow.graph import Graph
+
+
+def memory_chatbot_graph(template: str | None = None):
+    if template is None:
+        template = """{context}
+
+    User: {user_message}
+    AI: """
+    memory_component = MemoryComponent()
+    chat_input = ChatInput()
+    prompt_component = PromptComponent()
+    prompt_component.set(
+        template=template, user_message=chat_input.message_response, context=memory_component.retrieve_messages_as_text
+    )
+    openai_component = OpenAIModelComponent()
+    openai_component.set(input_value=prompt_component.build_prompt)
+
+    chat_output = ChatOutput()
+    chat_output.set(input_value=openai_component.text_response)
+
+    graph = Graph(chat_input, chat_output)
+    return graph

src/backend/base/langflow/initial_setup/starter_projects/sequential_tasks_agent.py

@@ -0,0 +1,91 @@
+from langflow.components.agents.CrewAIAgent import CrewAIAgentComponent
+from langflow.components.agents.SequentialCrew import SequentialCrewComponent
+from langflow.components.helpers.SequentialTask import SequentialTaskComponent
+from langflow.components.inputs.TextInput import TextInputComponent
+from langflow.components.models.OpenAIModel import OpenAIModelComponent
+from langflow.components.outputs.ChatOutput import ChatOutput
+from langflow.components.prompts.Prompt import PromptComponent
+from langflow.components.tools.SearchAPI import SearchAPIComponent
+from langflow.graph.graph.base import Graph
+
+
+def sequential_tasks_agent_graph():
+    llm = OpenAIModelComponent()
+    search_api_tool = SearchAPIComponent()
+    researcher_agent = CrewAIAgentComponent()
+    text_input = TextInputComponent(_display_name="Topic")
+    text_input.set(input_value="Agile")
+    researcher_agent.set(
+        tools=[search_api_tool.build_tool],
+        llm=llm.build_model,
+        role="Researcher",
+        goal="Search Google to find information to complete the task.",
+        backstory="Research has always been your thing. You can quickly find things on the web because of your skills.",
+    )
+    research_task = SequentialTaskComponent()
+    document_prompt_component = PromptComponent()
+    document_prompt_component.set(
+        template="""Topic: {topic}
+
+Build a document about this document.""",
+        topic=text_input.text_response,
+    )
+    research_task.set(
+        agent=researcher_agent.build_output,
+        task_description=document_prompt_component.build_prompt,
+        expected_output="Bullet points and small phrases about the research topic.",
+    )
+    editor_agent = CrewAIAgentComponent()
+    editor_task = SequentialTaskComponent()
+    revision_prompt_component = PromptComponent()
+    revision_prompt_component.set(
+        template="""Topic: {topic}
+
+Revise this document.""",
+        topic=text_input.text_response,
+    )
+    editor_agent.set(
+        llm=llm.build_model,
+        role="Editor",
+        goal="You should edit the Information provided by the Researcher to make it more palatable and to not contain misleading information.",
+        backstory="You are the editor of the most reputable journal in the world.",
+    )
+    editor_task.set(
+        agent=editor_agent.build_output,
+        task_description=revision_prompt_component.build_prompt,
+        expected_output="Small paragraphs and bullet points with the corrected content.",
+        task=research_task.build_task,
+    )
+    blog_prompt_component = PromptComponent()
+    blog_prompt_component.set(
+        template="""Topic: {topic}
+
+Build a fun blog post about this topic.""",
+        topic=text_input.text_response,
+    )
+    comedian_agent = CrewAIAgentComponent()
+    comedian_agent.set(
+        llm=llm.build_model,
+        role="Comedian",
+        goal="You write comedic content based on the information provided by the editor.",
+        backstory="Your formal occupation is Comedian-in-Chief. You write jokes, do standup comedy and write funny articles.",
+    )
+    blog_task = SequentialTaskComponent()
+    blog_task.set(
+        agent=comedian_agent.build_output,
+        task_description=blog_prompt_component.build_prompt,
+        expected_output="A small blog about the topic.",
+        task=editor_task.build_task,
+    )
+    sequential_crew_component = SequentialCrewComponent()
+    sequential_crew_component.set(tasks=blog_task.build_task)
+    chat_output = ChatOutput()
+    chat_output.set(input_value=sequential_crew_component.build_output)
+
+    graph = Graph(
+        start=text_input,
+        end=chat_output,
+        flow_name="Sequential Tasks Agent",
+        description="This Agent runs tasks in a predefined sequence.",
+    )
+    return graph

src/backend/base/langflow/initial_setup/starter_projects/vector_store_rag.py

@@ -0,0 +1,65 @@
+from textwrap import dedent
+
+from langflow.components.data.File import FileComponent
+from langflow.components.embeddings.OpenAIEmbeddings import OpenAIEmbeddingsComponent
+from langflow.components.helpers.ParseData import ParseDataComponent
+from langflow.components.helpers.SplitText import SplitTextComponent
+from langflow.components.inputs.ChatInput import ChatInput
+from langflow.components.models.OpenAIModel import OpenAIModelComponent
+from langflow.components.outputs.ChatOutput import ChatOutput
+from langflow.components.prompts.Prompt import PromptComponent
+from langflow.components.vectorstores.AstraDB import AstraVectorStoreComponent
+from langflow.graph.graph.base import Graph
+
+
+def ingestion_graph():
+    # Ingestion Graph
+    file_component = FileComponent()
+    text_splitter = SplitTextComponent()
+    text_splitter.set(data_inputs=file_component.load_file)
+    openai_embeddings = OpenAIEmbeddingsComponent()
+    vector_store = AstraVectorStoreComponent()
+    vector_store.set(
+        embedding=openai_embeddings.build_embeddings,
+        ingest_data=text_splitter.split_text,
+    )
+
+    ingestion_graph = Graph(file_component, vector_store)
+    return ingestion_graph
+
+
+def rag_graph():
+    # RAG Graph
+    openai_embeddings = OpenAIEmbeddingsComponent()
+    chat_input = ChatInput()
+    rag_vector_store = AstraVectorStoreComponent()
+    rag_vector_store.set(
+        search_input=chat_input.message_response,
+        embedding=openai_embeddings.build_embeddings,
+    )
+
+    parse_data = ParseDataComponent()
+    parse_data.set(data=rag_vector_store.search_documents)
+    prompt_component = PromptComponent()
+    prompt_component.set(
+        template=dedent("""Given the following context, answer the question.
+                         Context:{context}
+
+                         Question: {question}
+                         Answer:"""),
+        context=parse_data.parse_data,
+        question=chat_input.message_response,
+    )
+
+    openai_component = OpenAIModelComponent()
+    openai_component.set(input_value=prompt_component.build_prompt)
+
+    chat_output = ChatOutput()
+    chat_output.set(input_value=openai_component.text_response)
+
+    graph = Graph(start=chat_input, end=chat_output)
+    return graph
+
+
+def vector_store_rag_graph():
+    return ingestion_graph() + rag_graph()

src/backend/tests/unit/test_data_components.py

@@ -1,12 +1,11 @@
-import json
 import os
+import tempfile
 from pathlib import Path
 from unittest.mock import Mock, patch
 
 import httpx
 import pytest
 import respx
-from dictdiffer import diff
 from httpx import Response
 
 from langflow.components import data
@@ -167,23 +166,21 @@ def test_directory_component_build_with_multithreading(
 
 def test_directory_without_mocks():
     directory_component = data.DirectoryComponent()
-    from langflow.initial_setup import setup
-    from langflow.initial_setup.setup import load_starter_projects
 
-    _, projects = zip(*load_starter_projects())
-    # the setup module has a folder where the projects are stored
-    # the contents of that folder are in the projects variable
-    # the directory component can be used to load the projects
-    # and we can validate if the contents are the same as the projects variable
-    setup_path = Path(setup.__file__).parent / "starter_projects"
-    directory_component.set_attributes({"path": str(setup_path), "use_multithreading": False})
-    results = directory_component.load_directory()
-    assert len(results) == len(projects)
-    # each result is a Data that contains the content attribute
-    # each are dict that are exactly the same as one of the projects
-    for i, result in enumerate(results):
-        file_dict = json.loads(result.text)
-        assert file_dict in projects, list(diff(file_dict, projects[i]))
+    with tempfile.TemporaryDirectory() as temp_dir:
+        with open(temp_dir + "/test.txt", "w") as f:
+            f.write("test")
+        # also add a json file
+        with open(temp_dir + "/test.json", "w") as f:
+            f.write('{"test": "test"}')
+
+        directory_component.set_attributes({"path": str(temp_dir), "use_multithreading": False})
+        results = directory_component.load_directory()
+        assert len(results) == 2
+        values = ["test", '{"test":"test"}']
+        assert all(result.text in values for result in results), [
+            (len(result.text), len(val)) for result, val in zip(results, values)
+        ]
 
     # in ../docs/docs/components there are many mdx files
     # check if the directory component can load them