From 7c4c6b2a56efb48a7b5affeb4f7d1669bd767eaa Mon Sep 17 00:00:00 2001
From: Gabriel Luiz Freitas Almeida <gabriel@logspace.ai>
Date: Fri, 22 Mar 2024 22:37:26 -0300
Subject: [PATCH] Add XMLAgentComponent to build an XML agent from an LLM and
 tools

---
 .../langflow/components/agents/XMLAgent.py    | 90 +++++++++++++++++++
 1 file changed, 90 insertions(+)
 create mode 100644 src/backend/langflow/components/agents/XMLAgent.py

diff --git a/src/backend/langflow/components/agents/XMLAgent.py b/src/backend/langflow/components/agents/XMLAgent.py
new file mode 100644
index 000000000..9115b862f
--- /dev/null
+++ b/src/backend/langflow/components/agents/XMLAgent.py
@@ -0,0 +1,90 @@
+from typing import List
+
+from langchain.agents import AgentExecutor, create_xml_agent
+from langchain_core.prompts import PromptTemplate
+
+from langflow import CustomComponent
+from langflow.field_typing import BaseLLM, BaseMemory, Text, Tool
+
+
+class XMLAgentComponent(CustomComponent):
+    display_name = "XMLAgent"
+    description = "Construct an XML agent from an LLM and tools."
+
+    def build_config(self):
+        return {
+            "llm": {"display_name": "LLM"},
+            "tools": {"display_name": "Tools"},
+            "prompt": {
+                "display_name": "Prompt",
+                "multiline": True,
+                "info": "This prompt must contain 'tools' and 'agent_scratchpad' keys.",
+                "value": """You are a helpful assistant. Help the user answer any questions.
+
+            You have access to the following tools:
+
+            {tools}
+
+            In order to use a tool, you can use <tool></tool> and <tool_input></tool_input> tags. You will then get back a response in the form <observation></observation>
+            For example, if you have a tool called 'search' that could run a google search, in order to search for the weather in SF you would respond:
+
+            <tool>search</tool><tool_input>weather in SF</tool_input>
+            <observation>64 degrees</observation>
+
+            When you are done, respond with a final answer between <final_answer></final_answer>. For example:
+
+            <final_answer>The weather in SF is 64 degrees</final_answer>
+
+            Begin!
+
+            Previous Conversation:
+            {chat_history}
+
+            Question: {input}
+            {agent_scratchpad}""",
+            },
+            "tool_template": {
+                "display_name": "Tool Template",
+                "info": "Template for rendering tools in the prompt. Tools have 'name' and 'description' keys.",
+                "advanced": True,
+            },
+            "handle_parsing_errors": {
+                "display_name": "Handle Parsing Errors",
+                "info": "If True, the agent will handle parsing errors. If False, the agent will raise an error.",
+                "advanced": True,
+            },
+            "memory": {
+                "display_name": "Memory",
+                "info": "Memory to use for the agent.",
+            },
+        }
+
+    def build(
+        self,
+        inputs: str,
+        llm: BaseLLM,
+        tools: List[Tool],
+        prompt: str,
+        memory: BaseMemory = None,
+        tool_template: str = "{name}: {description}",
+        handle_parsing_errors: bool = True,
+    ) -> Text:
+        if "input" not in prompt:
+            raise ValueError("Prompt must contain 'input' key.")
+
+        def render_tool_description(tools):
+            return "\n".join([tool_template.format(name=tool.name, description=tool.description) for tool in tools])
+
+        prompt_template = PromptTemplate.from_template(prompt)
+        input_variables = prompt_template.input_variables
+        agent = create_xml_agent(llm, tools, prompt_template, tools_renderer=render_tool_description)
+        runnable = AgentExecutor.from_agent_and_tools(
+            agent=agent, tools=tools, verbose=True, memory=memory, handle_parsing_errors=handle_parsing_errors
+        )
+        input_dict = {"input": inputs}
+        for var in input_variables:
+            if var not in ["agent_scratchpad", "input"]:
+                input_dict[var] = ""
+        result = runnable.invoke(input_dict)
+        self.status = result
+        return result