beam-langchain

LangChain integration for the Beam Protocol.

This package bridges LangChain tools with Beam's agent-to-agent messaging model so an LLM can:

• send natural-language messages to remote Beam agents with conversation.message
• expose Beam agent capabilities as LangChain tools
• use the hosted Beam directory at https://api.beam.directory

Installation

pip install beam-langchain

Quick Start

import asyncio

from beam_directory import BeamClient, BeamIdentity
from beam_langchain import BeamAgentTool, BeamToolkit


async def main() -> None:
    identity = BeamIdentity.generate(agent_name="planner", org_name="demo")
    client = BeamClient(identity=identity, directory_url="https://api.beam.directory")

    message_tool = BeamAgentTool(
        client=client,
        beam_id="researcher@demo.beam.directory",
        name="beam_researcher_message",
        description="Ask the remote research agent for a natural-language answer.",
    )

    reply = await message_tool._arun(message="Summarize today's support queue.")
    print(reply)

    toolkit = await BeamToolkit.afrom_agents(
        client,
        ["researcher@demo.beam.directory"],
    )
    for tool in toolkit.get_tools():
        print(tool.name)


asyncio.run(main())

API

BeamAgentTool

BeamAgentTool wraps Beam's conversation.message intent as a LangChain tool.

tool = BeamAgentTool(
    client=client,
    beam_id="support@demo.beam.directory",
)

answer = await tool._arun(
    message="What incidents are still open?",
    context={"priority": "high"},
)

BeamToolkit

BeamToolkit converts Beam agents and their advertised capabilities into LangChain tools.

from beam_directory.types import AgentSearchQuery

toolkit = await BeamToolkit.afrom_search(
    client,
    AgentSearchQuery(org="demo", limit=5),
)

tools = toolkit.get_tools()

Each discovered agent produces:

• one BeamAgentTool for conversation.message
• one intent tool per declared Beam capability

Development

cd packages/beam-langchain
python -m venv .venv
source .venv/bin/activate
pip install -e .
python -m unittest discover -s tests

License

Apache 2.0 — see ../../LICENSE.