Framework

BeamAI Extra · Tools, RAG, and protocol extensions for BeamAI

The BeamAI core handles only the agent foundations; real deployments need tools, retrieval, and protocols for the outside world. BeamAI Extra packages those as five independent OTP applications — pull in whichever one you use.

It depends on the main branch of the BeamAI core and shares its ChatClient, Tool, and Filter conventions.

Tools · RAG · MCP · A2A · AG-UI

Erlang: create and run an agent
LLM = beamai_chat_model:create(anthropic, #{
    model => <<"glm-4.7">>,
    api_key => list_to_binary(os:getenv("ZHIPU_API_KEY")),
    base_url => <<"https://open.bigmodel.cn/api/anthropic">>
}),

{ok, Agent0} = beamai_agent:new(#{
    system_prompt => <<"You are a helpful assistant."/utf8>>,
    llm => LLM
}),

{ok, Result, _Agent1} = beamai_agent:run(Agent0, <<"Hello!"/utf8>>),
Content = maps:get(content, Result).

Features

What it can do

  • beamai_tools: tools and middleware

    Ready-made file, shell, todo, and human-interaction tools, plus a middleware system that wraps tool calls with permissions, confirmation, and audit.

  • beamai_rag: retrieval-augmented generation

    Text splitting, embedding, and similarity search — bring your own material into the agent's context.

  • beamai_mcp: Model Context Protocol

    An MCP client and server: let agents use external MCP tools, and expose your own capabilities as an MCP service.

  • beamai_a2a: Agent-to-Agent

    The A2A protocol implemented, so agents in different processes and different systems can discover and call each other.

  • beamai_agui: AG-UI runtime

    A server for agent event streams, modeled on the CopilotKit runtime, pushing agent progress to a frontend as AG-UI events.

Specifications

Technical specifications

RuntimeErlang/OTP, built with rebar3; depends on the BeamAI core (main branch)
Modulesbeamai_tools, beamai_rag, beamai_mcp, beamai_a2a, beamai_agui — five OTP applications
LicenseApache License 2.0

FAQ

You might be wondering

Do I need all five extensions?

No. The five extensions are independent OTP applications — depend only on the one you need.

MCP client, server, or both?

Both: agents can call tools on external MCP services, and your system can serve its own tools over MCP.

What is the license?

Apache License 2.0, fine for commercial projects.

Open source

Apache 2.0 open source

The engine, tests, benchmarks, and design docs are all in the repository — read them, change them, file issues.

Related framework

BeamAI · The AI agent framework for Erlang/OTP

Build AI agents on Erlang/OTP: a unified LLM client across six vendors, ChatClient and tool calling, onion-style filters, and a ReAct agent with conversation memory.