BeeAI Framework: AI agents with rule-guided behavior
BeeAI Framework is an open source library for building AI agents and multi-agent systems in Python or TypeScript. Created by IBM and later handed to the Linux Foundation, it targets developers who want predictable agents regardless of the model behind them.
Data as of October 22, 2026
Key figures
| Developer | Projet BeeAI (LF AI & Data, Linux Foundation), créé et contribué par IBM |
|---|---|
| Type | Code |
| Languages | Python, TypeScript |
| Open source | Yes |
| License | Apache-2.0 |
| First release | 2024 |
| Suggested level | Intermediate |
| Use cases | multi-agent, workflows, RAG |
| Self-hostable | Yes |
| Business model | Free |
| Cloud offering | No |
| Multi-agent | Yes |
| MCP support | Yes |
| Bring your own model | Yes |
| Official site | Official site · Documentation · Repository |
GitHub
| Repository | i-am-bee/beeai-framework |
|---|---|
| GitHub stars | 3,439 |
| Forks | 535 |
| Latest code activity | October 21, 2026 |
| Latest release | typescript_v0.1.31 (September 28, 2026) |
GitHub activity checked on October 22, 2026.
From IBM to the Linux Foundation
BeeAI started at IBM, first in TypeScript, then with a Python library released as an alpha in February 2025. In April 2025 IBM contributed it to the LF AI & Data Foundation along with two other projects. The framework now aims for full feature parity between Python and TypeScript, which is unusual: most competitors favor one language.
What you can build with it
BeeAI provides the standard building blocks of an AI agent: tools (web search, weather, code execution or your own), conversation memory, caching, state serialization, RAG with vector stores, prompt templates, and “workflows” for orchestrating several agents. Its signature feature is the Requirement Agent, which lets you set rules the agent must follow, reducing unpredictable behavior from one model to the next.
For interoperability, BeeAI supports MCP for external tools and the A2A protocol for talking to agents built with other frameworks. A “Serve” module exposes your agents on a server through these protocols.
Getting started
The easiest path is one of the official starter templates (Python or TypeScript), or installing the beeai-framework package with pip or npm. Install Ollama and a Granite model to run the examples for free on your own machine, then switch to the provider you prefer. The reference example shows a main agent handing questions off to two specialist agents.
Strengths and limitations
Strengths: two languages on an equal footing, rule-guided agents that behave more predictably, built-in MCP and A2A, neutral governance and an Apache 2.0 license.
Limitations: the community and tutorial ecosystem are smaller than those of the best-known frameworks; released versions are still 0.x, a sign the API may keep changing; and IBM does not guarantee support. For business use, keep a close eye on release notes.
Alternatives
CrewAI and LangGraph are more widely used for multi-agent work, Strands Agents also covers Python and TypeScript, and Mastra is a TypeScript-first option.
Frameworks of the same type
Frequently asked questions
Is BeeAI an IBM product?
IBM created it and donated it to the LF AI & Data Foundation, part of the Linux Foundation, in April 2025. It has followed open governance since then. The repository's legal notice states that IBM is not committed to providing support, so treat it as a community project rather than a commercial product.
Which models work with BeeAI?
The framework uses a shared backend layer that connects to several providers, including Ollama, Groq, OpenAI and watsonx.ai. The official examples run a Granite model locally through Ollama, so you can experiment without paying for an API.
What is the Requirement Agent?
It's an agent type you can give explicit rules to, such as thinking before acting or calling a specific tool at a given step. The goal is more consistent behavior, even when you swap models.
Is there a paid hosted version?
No. The repository and documentation do not mention any paid cloud offering for the framework. You host it yourself; only calls to a remote model are billed, by that model's provider.
Data sources
- github.com
- framework.beeai.dev
- linuxfoundation.org
- GitHub — Repository metadata: GitHub
Data as of October 22, 2026
Report an error
Does a figure, a date or a fact look wrong? Tell us: every report is checked.