LangChain Integration

buddy.integrations.langchain bridges Buddy and the LangChain ecosystem in both directions.

pip install "buddy-ai[langchain]"

from buddy.integrations.langchain import (
    BuddyChatModel,
    BuddyAgentTool,
    to_langchain_tool,
    from_langchain_tool,
    to_buddy_messages,
    from_buddy_message,
)

Buddy model → LangChain chat model

BuddyChatModel wraps any buddy.models model as a LangChain BaseChatModel. This means Buddy's unified provider layer (OpenAI, Anthropic, Google, Cohere, AWS, Azure, Ollama, Groq, …) can drive LangChain chains, agents, and LangGraph nodes — one model object, every provider.

from buddy.models.openai import OpenAIChat
from buddy.integrations.langchain import BuddyChatModel

llm = BuddyChatModel(buddy_model=OpenAIChat(id="gpt-4o-mini"))

# Standard LangChain Runnable interface
print(llm.invoke("Say hello in French.").content)

# Works in LCEL chains
from langchain_core.prompts import ChatPromptTemplate

prompt = ChatPromptTemplate.from_messages([("human", "Translate to {lang}: {text}")])
chain = prompt | llm
print(chain.invoke({"lang": "German", "text": "good morning"}).content)

Both sync (invoke) and async (ainvoke) are supported; _agenerate calls the Buddy model's aresponse.

How it works

BuddyChatModel is a lightweight proxy until first construction, so importing the module never requires LangChain. On first use it builds a real BaseChatModel subclass against your installed langchain-core, converts LangChain messages to Buddy Message objects, calls model.response(...), and returns a ChatResult.

Buddy agent → LangChain tool

BuddyAgentTool exposes a full Buddy Agent or Team — with its own model, tools, memory, and knowledge — as a single LangChain tool that other LangChain agents can call.

from buddy import Agent
from buddy.models.openai import OpenAIChat
from buddy.integrations.langchain import BuddyAgentTool

researcher = Agent(name="researcher", model=OpenAIChat(id="gpt-4o-mini"))

research_tool = BuddyAgentTool(
    researcher,
    name="research",
    description="Research a topic and return concise findings.",
).as_tool()

# `research_tool` is a langchain_core StructuredTool (sync + async).

Tool conversion (both directions)

Buddy → LangChain

from buddy.tools import tool
from buddy.integrations.langchain import to_langchain_tool

@tool
def word_count(text: str) -> int:
    """Count the number of words in a piece of text."""
    return len(text.split())

lc_tool = to_langchain_tool(word_count)   # a StructuredTool
lc_tool.invoke({"text": "one two three"})  # -> 3

to_langchain_tool accepts a Buddy Function (from @tool) or any plain callable; callables are wrapped via Function.from_callable.

LangChain → Buddy

from buddy import Agent
from buddy.models.openai import OpenAIChat
from buddy.integrations.langchain import from_langchain_tool

buddy_fn = from_langchain_tool(lc_tool)   # a buddy Function

agent = Agent(model=OpenAIChat(id="gpt-4o-mini"), tools=[buddy_fn])

The converter extracts the tool's argument schema (pydantic v1 or v2 args_schema) into Buddy's JSON-schema parameter format, and delegates execution to the LangChain tool's run method.

Message conversion

from langchain_core.messages import HumanMessage, SystemMessage
from buddy.integrations.langchain import to_buddy_messages, from_buddy_message
from buddy.models.message import Message

buddy_msgs = to_buddy_messages([SystemMessage(content="..."), HumanMessage(content="hi")])
lc_msg = from_buddy_message(Message(role="assistant", content="hello"))

Role mapping: human↔user, ai↔assistant, system↔system, tool↔tool (preserving tool_call_id).

API summary

Symbol	Type	Description
`BuddyChatModel(buddy_model=...)`	factory	LangChain `BaseChatModel` backed by a Buddy model
`BuddyAgentTool(agent, name=, description=).as_tool()`	class	Buddy agent/team as a LangChain tool
`to_langchain_tool(fn)`	function	Buddy `Function`/callable → LangChain `StructuredTool`
`from_langchain_tool(tool)`	function	LangChain `BaseTool` → Buddy `Function`
`to_buddy_messages(messages)`	function	LangChain messages → Buddy `Message` list
`from_buddy_message(message)`	function	Buddy `Message` → LangChain message