Beige is a secure, open-source, sandboxed agent system where AI agents write and execute code inside Docker containers. The gateway orchestrates LLM calls, enforces policies, audit-logs every tool invocation, and routes tool execution.

Documentation — Full docs at beige.mintlify.app

Key Features

What's common to tools like OpenClaw

• Channels — Interact via TUI, Telegram, HTTP API, or custom channels
• Skills — Read-only knowledge packages for agent context
• Tools — Extensible tool system for agent capabilities
• Policy Enforcement — Fine-grained control over what agents can do
• Audit Logging — Every tool invocation is logged for accountability

What's different

• Tools become CLI executables — Agents write scripts that call tools directly, eliminating round-trips through the LLM
• Docker Sandboxing — always — All code execution happens in isolated containers with no escape hatch. Communication to the gateway via sockets.
• Nothing out-of-the-box — You configure agents the way YOU want

Why Beige? {#why-beige}

True Autonomy {#true-autonomy}

Traditional tool-calling requires the LLM to invoke tools one at a time. Each result goes back through the model, wasting tokens and time. Complex workflows require dozens of individual tool calls.

Beige agents can write and run code. Instead of calling a tool 20 times, the agent writes a script that does it in a loop. The LLM only sees the final result.

| Beige | Traditional LLM | | ------------------------------------------------------------------ | ---------------------------------------------------------- | | 1 round-trip — agent writes a script, runs it, gets one result | 20 round-trips — each tool call goes through the model | | 20 KV lookups happen inside the sandbox | 20× the latency, 20× the token overhead |

Beige — write then exec

The agent writes a TypeScript file, then executes it. One round-trip to the LLM.

const users = [];
for (let i = 1; i <= 20; i++) {
  const result = await exec(`/tools/bin/kv get user:${i}`);
  users.push(JSON.parse(result));
}
console.log(JSON.stringify(users));

write /workspace/fetch-users.ts
exec deno run /workspace/fetch-users.ts

Result returned to LLM: one JSON array — done.

Every tool call — including the ones called by a script — is routed through the gateway. The gateway checks permissions, enforces policy, and ensures secrets never reach the sandbox.

Traditional LLM

Without a code runtime the LLM must call the tool once per item. Slow, unsafe, error-prone, not easily reproducible:

→ tool_call: kv get user:1
← result: {"id":1,"name":"Alice"}
→ tool_call: kv get user:2
← result: {"id":2,"name":"Bob"}
... 18 more calls ...

Quick Start

Prerequisites: Node.js 22+, Docker

Install Beige globally:

npm install -g @matthias-hausberger/beige

Set your Anthropic API key (you can also set this in your ~/.beige/config.json5 config after you started the gateway for the first time):

export ANTHROPIC_API_KEY="sk-ant-..."

Start the gateway:

beige gateway start

Open the TUI. Alternatively you can also use Telegram or add your own channel plugin:

beige tui

On first run, Beige creates ~/.beige/ with a default config and the bundled KV tool.

You can also use any other LLM Provider.

See the installation guide for details.

Documentation

Full documentation at beige.mintlify.app:

| Section | Description | | ------------------------------------------------------------------- | ------------------------------------------- | | Getting Started | Install and run your first agent | | Gateway | Core concepts, tool calls, operations | | Agents | Configure providers, models, tools, skills | | Channels | TUI, Telegram, HTTP API | | Tools | Core tools, building custom tools, toolkits | | Config Reference | Complete config file reference | | CLI Reference | All CLI commands and flags |

Contributing

All contributions are welcome. See CONTRIBUTING.md for development setup.

For source installs, Beige runs entirely inside the repo — no files are written to your home directory.

Clone and install:

git clone https://github.com/matthias-hausberger/beige.git
cd beige
pnpm install

Run setup:

pnpm run beige setup

This sets BEIGE_HOME=./.beige automatically, so the repo is self-contained.

License

MIT