golembot

English | 中文

Cursor, Claude Code, OpenCode, Codex — these Coding Agents can already write code, run scripts, analyze data, and reason through complex tasks. But they're stuck in an IDE or a terminal window.

GolemBot gives them a body. One command connects your Coding Agent to Slack, Telegram, Discord, Feishu, DingTalk, WeCom, WeChat, or any HTTP client — with any LLM provider. Route Claude Code through OpenRouter, run Codex on MiniMax, or point OpenCode at DeepSeek — one config block, zero code changes. Write a custom adapter to plug in email, GitHub Issues, or any other message source. Or embed into your own product with 5 lines of code. No AI framework, no prompt engineering — the agent you already have is the brain.

Run Your Coding Agent Everywhere

On IM — your team's 24/7 AI teammate

golembot init -e claude-code -n my-bot
golembot gateway    # Slack, Telegram, Discord, Feishu, DingTalk, WeCom, WeChat

Your colleagues @ the bot in group chat. It can write code, analyze files, answer questions — because behind it is a real Coding Agent, not a thin API wrapper.

In your product — full agent power, 5 lines of code

import { createAssistant } from 'golembot';
const bot = createAssistant({ dir: './my-agent' });

for await (const event of bot.chat('Analyze last month sales data')) {
  if (event.type === 'text') process.stdout.write(event.content);
}

Embed into Slack bots, internal tools, SaaS products, customer support — anything that speaks Node.js.

Why GolemBot, not another AI framework?

Quick Start

npm install -g golembot

mkdir my-bot && cd my-bot
golembot onboard      # guided setup (recommended)

# Or manually:
golembot init -e claude-code -n my-bot
golembot run          # REPL conversation
golembot gateway      # start IM + HTTP service + Dashboard
golembot fleet ls     # list all running bots
golembot skill search "data analysis"  # browse 13,000+ ClawHub skills

Long task still running? Use /stop in REPL or IM, assistant.cancel(sessionKey?) in code, or POST /abort over HTTP to cancel the current task without clearing session history.

Dashboard & Fleet

Every golembot gateway instance comes with a built-in web Dashboard showing real-time metrics, channel status, and a quick-test console:

Running multiple bots? golembot fleet serve aggregates them into a single Fleet Dashboard:

Architecture

Slack / Telegram / Discord / Feishu / DingTalk / WeCom / WeChat / HTTP API
    Custom Adapters (email, GitHub Issues, ...)
                    │
                    ▼
         ┌─────────────────────────┐
         │     Gateway Service     │
         │  (Channel adapters +    │
         │   HTTP service)         │
         └────────────┬────────────┘
                      │
              createAssistant()
                      │
          ┌───────┬───────┬───────┐
          ▼       ▼       ▼       ▼
       Cursor  Claude  OpenCode  Codex
               Code
          ↕ Provider Routing (OpenRouter, MiniMax, ...)

Engine Comparison

The StreamEvent interface is identical across all engines — switching requires zero code changes.

If you run Codex, note that GolemBot defaults to codex.mode: unrestricted. Set codex.mode: safe if you want to keep Codex sandboxed. You can also use fine-grained 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 expose /chat/completions or Anthropic-style /messages endpoints will fail. See the Provider Routing guide: https://0xranx.github.io/golembot/guide/provider-routing#codex-requires-responses-api

Configuration

golem.yaml — the single config file:

name: my-assistant
engine: claude-code

# Optional: route engine to a third-party LLM provider
provider:
  baseUrl: "https://openrouter.ai/api"
  apiKey: "${OPENROUTER_API_KEY}"
  model: "anthropic/claude-sonnet-4"

channels:
  slack:
    botToken: ${SLACK_BOT_TOKEN}
    appToken: ${SLACK_APP_TOKEN}
  telegram:
    botToken: ${TELEGRAM_BOT_TOKEN}
  discord:
    botToken: ${DISCORD_BOT_TOKEN}
    botName: my-assistant        # optional — normalizes @mention tokens in channel messages
  feishu:
    appId: ${FEISHU_APP_ID}
    appSecret: ${FEISHU_APP_SECRET}
    # domain: lark                 # optional — Lark global tenants
    # pingTimeout: 30              # optional — WebSocket pong timeout in seconds
  # Custom adapter — local file or npm package
  my-email:
    _adapter: ./adapters/email-adapter.js
    token: ${EMAIL_TOKEN}

# Optional: Codex runtime controls (Codex engine only)
codex:
  mode: unrestricted   # compatibility alias; set safe to keep sandboxing
  sandbox: workspace-write
  approval: on-request
  search: true
  addDirs:
    - ../shared-assets

gateway:
  port: 3000
  token: ${GOLEM_TOKEN}

Sensitive fields support ${ENV_VAR} references. Custom channel adapters can be local .js/.mjs files or npm packages — see the adapter guide.

Skill System

A Skill is a directory containing SKILL.md + optional scripts. Drop it in, the assistant gains new abilities. Remove it, the ability is gone.

skills/
├── general/          # Built-in: general assistant
│   └── SKILL.md
├── im-adapter/       # Built-in: IM reply conventions
│   └── SKILL.md
└── my-custom-skill/  # Your own
    ├── SKILL.md
    └── analyze.py