GolemBot's positioning is "Any Agent × Any Provider × Anywhere". The core problem it solves is: coding agents such as Cursor, Claude Code, OpenCode, and Codex are already capable of writing code, running scripts, analyzing data, and reasoning about complex tasks, but they are usually confined to an IDE or terminal window; GolemBot gives them a "body" to connect with the outside world.
Main capabilities
- Multi-channel connectivity: with a single command, connect coding agents to Slack, Telegram, Discord, Feishu, DingTalk, WeCom, WeChat, or any HTTP client.
- Any LLM provider: let Claude Code go through OpenRouter, run Codex on MiniMax, or point OpenCode at DeepSeek, all through one configuration block, with no code changes.
- Custom adapters: write custom adapters to connect to email, GitHub Issues, or other message sources.
- Embed in your own product: embed into a Slack bot, internal tool, SaaS product, or customer support system with about 5 lines of code (any scenario that supports Node.js).
- No dependency on AI frameworks or prompt engineering: use your existing agent as the "brain".
Usage
- In IM: colleagues can simply @ the bot in a group chat, and the real coding agent behind it handles tasks such as writing code, analyzing files, and answering questions.
- In products: embed with a small amount of code to get full agent capabilities.
Task control
- When a long-running task is still running, you can use /stop in the REPL or IM, use assistant.cancel(sessionKey?) in code, or send POST /abort over HTTP to cancel the current task without clearing the session history.
Dashboard and Fleet
- Every golembot gateway instance has a built-in Web Dashboard showing real-time metrics, channel status, and a quick test console.
- When running multiple bots, you can use golembot fleet serve to aggregate them into a Fleet Dashboard.
Engine and configuration
- All engines have a consistent StreamEvent interface, so switching engines requires noModify the code.
- When using Codex, GolemBot defaults to codex.mode: unrestricted; if you need to keep the Codex sandbox, you can set codex.mode: safe, or use the more granular codex.sandbox, codex.approval, codex.search, and codex.addDirs settings.
- If you route Codex through a custom provider, that provider must support the OpenAI Responses API; providers that only offer /chat/completions or Anthropic-style /messages endpoints will fail.
- A single configuration file, golem.yaml; sensitive fields support ${ENV_VAR} references; custom channel adapters can be local .js / .mjs files or npm packages.
Skill System
- A Skill is a directory containing SKILL.md and optional scripts; drop it in to gain a new capability, remove it and the capability disappears.
- ls skills/ lists everything the assistant can currently do.
- Fully compatible with ClawHub (OpenClaw's AI agent skill marketplace), the SKILL.md format is 100% compatible, and 13,000+ community skills work out of the box.
- Supports agent-driven skill discovery: have the agent search for and install skills in conversation, for example asking "find me a good code review skill", and it will search ClawHub, show the results, and install after confirmation.
- All skill commands support --json for programmatic access; a pluggable registry interface supports additional skill sources beyond ClawHub.
Deployment
- Supports Docker deployment.
- Contributions are welcome; see CONTRIBUTING.md for contribution guidelines.
Limitations and Notes
- The material does not provide specific pricing, prices, platform support details, or performance data.
- Details regarding Codex provider restrictions, safe modes, etc. should be based on the official Provider Routing guide.
