Ie                á”
                     (ÆÆ           ÌÆÆ­
                        ÆÆ        ÆÓ
  jÆÆÆÞ             ÆÆ    ÆÆÆä1ÆÆè    ÆÆ              ;‹
  ’„‹ ÆÆ=ÆÆ          Æ    BÆÆÆÆÆÆ    ÆW          ÆÆÆÆÆì»Í~
   îÍ  ?RÆAÆÆÆÆM     ÍÆ  ÆÆ  ‹Ò ÆÆ  ÆÆ       ÆÆƉ ñ    ¯ç–
    «¤’  Æ  å„ «ÆÆÆ    Æ ÆÆÆÆÆÆÆÆÆ ÆÖ    ÆÆÆ®õwÆ ÆP í–¯ª
      íÉ%  +  ¼º  ÛÞÆÆ  Y  üÆÆÆå‚ Æ   ÆÆÆÆ  ;Ø  ú   ïº:
         oÑQÆÆÆÆÆØÉÝÅBÆÆÆÆÇ-BEE-ÆÆÆÆÆÆÆÑ  xŸÁÆÆz®ç†
         ¾Ÿ}sÆÆÞâÆÆÆÆÆÆÆÊÆ-AGENT-7Ëà          ’ÆðF
                   ˜      ÇÆÆÆÆÆØZ    ¾ÆÑÆQ˧¼÷
                      ÆÆÆÆ ôXÆÆã“sÇÆÆÑ
                   ÆÆÆ    #ÆÄiSœÆÆ    ÆÆÆ
                 ÆÆ  ÆÆTòHÆÆÆÆÆÆÆÆXiÆÆÆ  ÆÆ
               ÆÆ  ÆÆÍ  l=¡5ÇÇS9¾yï×  —Æ  FÆ
             ÆÆ   ÆÆ   ¨ÆÆÆÆÆÆÆÆÆÆÆÆ}  ÆÆ   ÆÆ
            Æ4   3Ʋ   »ÇÇoàÇSL…¹ ÇÇ    ÆÆ    Ʀ
                 ÆÆ     GÆÆÆÆÆÆÆÆÆÆÒ7   ËÆ
                 Ư      ÆÆæëÚšYJQÄÇ     Æ
                1Æ       çØÆÆÆÆÆÆÆÇ      Æ6
               ÆÆ          §ËÆÆÆé         Ƴ
             –ÆÆ                           Æ

bee

tiny but mighty

agent harness for local models

bee agent screenshot bee agent screenshot bee agent screenshot bee agent screenshot bee agent screenshot

Install

curl -fsSL https://raw.githubusercontent.com/elhenro/bee/main/install.sh | sh

or: go install github.com/elhenro/bee/cmd/bee@latest

then run bee, type /model, choose oMLX, Ollama, OpenRouter, etc. and pick a model. Local or hosted, your choice.

What is it?

🐝 Coding agent — writes code, runs tests, commits changes
⚡ Pure Go — single static binary, no runtime deps
🍳 Skills — `bee <name>` subcommands, one binary, one PATH
💻 Works everywhere — Ollama local to OpenRouter, tiny models to frontier
👑 Three modes — worker (full surface), scout (read-only research), queen (hive planning)
🤝 Handoff — stuck small model? hand off to a bigger one with a rescue brief

Local LLMs

bee is built to work well with Ollama and oMLX so you can run locally on your own hardware and keep full control. No API keys, no rate limits, no data leaving your machine.

On macOS, oMLX works best. Native Apple Silicon acceleration with prompt caching keeps things fast and memory-efficient.

I run Huihui-Qwen3.6-35B-A3B-Claude-4.7-Opus-abliterated-mlx-8bit (34.32 GB) very reliably on a MacBook M3 Max 64 GB, with Qwen3-VL-4B-Instruct-MLX-4bit (2.90 GB) for vision support. bee handles vision automatically for models that only do text.

Settings for the Huihui-Qwen model: temperature 0.7, top-p 0.85, top-k 20, KV-cache quantization 8-bit.

Other models that work well:

gemma-4-12B-it-4bit — 10.26 GB
gemma-4-12B-it-8bit — 11.87 GB
gemma-4-12B-it-assistant-bf16 — 837 MB
Qwen3-Coder-Next-4bit — 41.78 GB
Qwen3.6-27B-8bit — 9.02 GB
Qwen3.6-35B-A3B-4bit — 19.03 GB
Qwen3.6-35B-A3B-8bit — 29.68 GB

FAQ

What is bee?

bee is a minimal coding agent harness written in Go. It is a single static binary that turns any LLM into an autonomous agent which writes code, runs tests, and commits changes. There is no runtime to install and no daemon, just one executable on your PATH.

How do I install bee?

Run curl -fsSL https://raw.githubusercontent.com/elhenro/bee/main/install.sh | sh, or go install github.com/elhenro/bee/cmd/bee@latest if you have Go. Then run bee, type /model, choose a runtime such as oMLX, Ollama, or OpenRouter, and pick a model.

Can bee run fully local and offline?

Yes. bee runs against local models through Ollama or oMLX, so there are no API keys, no rate limits, and nothing leaves your machine. On macOS, oMLX adds native Apple Silicon acceleration with prompt caching for fast, memory-efficient inference.

What hardware do I need to run bee locally?

It scales to your machine. A MacBook M3 Max with 64 GB of RAM reliably runs a 34 GB 8-bit Qwen3.6-35B model. Smaller models such as gemma-4-12B at 10 GB run on far less, so a modern laptop with 16 to 32 GB is enough to start.

Which models work well with bee?

bee works from tiny local models to frontier APIs. The reference model is Huihui-Qwen3.6-35B-A3B at 8-bit, with Qwen3-VL-4B for vision; gemma-4-12B and Qwen3-Coder-Next also perform well. bee handles vision automatically for models that only do text.

What are bee skills?

Skills are extra capabilities exposed as subcommands. You invoke them as bee <name>, so one binary on your PATH covers many jobs instead of a directory of separate scripts. This keeps installation to a single file while letting bee grow new abilities.

Does bee work with hosted providers like OpenRouter?

Yes. Besides local Ollama and oMLX, bee connects to hosted providers such as OpenRouter, so you can mix local and cloud models. Switch at any time with the /model command and pick whichever runtime and model fit the task and your budget.

Is bee free and open source?

Yes. bee is free and open source, with the full source, issues, and releases on GitHub at github.com/elhenro/bee. It is pure Go and ships as a single static binary for macOS, Linux, and Windows, with no runtime dependencies to manage.

Get involved

GitHub Releases