golem

Golem (גּוֹלֶם)

Golem is a personal AI assistant that lives in your terminal and works while you sleep. Built with Go and Eino, it connects to any major LLM provider, executes shell commands, manages files, searches the web, and runs scheduled tasks — all from a single binary with zero external dependencies.

Golem (גּוֹלֶם): In Jewish folklore, a Golem is an animated being created from inanimate matter — an obedient servant that tirelessly performs tasks for its creator.

中文文档

---

Why Golem?

• One binary, zero bloat. No Python, no Node, no Docker. Just a single Go binary.
• Provider-agnostic. Switch between 9 LLM providers with a config change — OpenRouter, Claude, OpenAI, DeepSeek, Gemini, Ark, Qianfan, Qwen, or Ollama.
• Always-on option. Run as a background server with multi-channel integration (Telegram, WhatsApp, Feishu, Discord, Slack, QQ, DingTalk, MaixCam), HTTP gateway, and cron-scheduled tasks.
• Tool-wielding agent. Not just a chatbot — Golem reads files, runs commands, searches the web, and remembers context across sessions.
• Extensible by design. Install skill packs from GitHub to teach it new tricks.

---

Architecture

┌──────────────────────────────────────────────────┐
│              Channels (Input/Output)             │
│         CLI TUI  ·  Telegram  ·  Gateway API     │
└────────────────────────┬─────────────────────────┘
                         │
              ┌──────────▼──────────┐
              │     Message Bus     │
              │  (event-driven, Go  │
              │      channels)      │
              └──────────┬──────────┘
                         │
              ┌──────────▼──────────┐
              │     Agent Loop      │
              │ (iterative LLM +    │
              │   tool calling)     │
              └──────────┬──────────┘
                         │
        ┌────────────────┼────────────────┐
        │                │                │
   ┌────▼────┐    ┌──────▼──────┐   ┌────▼────┐
   │   LLM   │    │    Tools    │   │ Storage │
   │Providers│    │             │   │         │
   └─────────┘    └─────────────┘   └─────────┘
   OpenRouter      exec              Sessions
   Claude          read_file         (JSONL)
   OpenAI          write_file        Memory
   DeepSeek        list_dir          Skills
   Gemini          read_memory       Cron Jobs
   Ark             write_memory
   Qianfan         append_diary
   Qwen            web_search
   Ollama          web_fetch
                   manage_cron

---

✨ Features

🖥️ Interactive Terminal UI

A rich chat experience powered by Bubble Tea — autocomplete, streaming responses, and full tool execution, all in your terminal.

🤖 Server Mode

Run golem run as a background service. Connects enabled channels, the HTTP Gateway API, and the cron scheduler simultaneously.

🛠️ 10 Built-in Tools

Tool	What it does
exec	Run shell commands (sandboxing available)
read_file / write_file	Read and write files in the workspace
list_dir	Browse directory contents
read_memory / write_memory	Persistent long-term memory
append_diary	Daily diary notes
web_search	Search the web (Brave when API key exists, otherwise free DuckDuckGo fallback)
web_fetch	Fetch and extract web page content
manage_cron	Create and manage scheduled tasks

🔌 9 LLM Providers

OpenRouter · Claude (Anthropic) · OpenAI · DeepSeek · Gemini · Ark · Qianfan · Qwen · Ollama — all via a unified OpenAI-compatible interface.

⏰ Cron Scheduler

Persistent, auto-resuming scheduled tasks. Supports one-shot (at), interval (every), and cron expressions. The agent can even schedule its own tasks.

🧩 Skills System

Install markdown-based skill packs from GitHub to extend the agent's system prompt with domain-specific instructions.

📡 Gateway API

HTTP endpoints (/health, /version, /chat) available during golem run for programmatic access from external services.

---

Installation

Download Binary (Recommended)

Grab the pre-compiled binary for Windows or Linux from the Releases page.

Install from Source

go install github.com/MEKXH/golem/cmd/golem@latest

---

Quick Start

1. Initialize — generates ~/.golem/config.json:

golem init

2. Add your API key — edit ~/.golem/config.json:

{
  "agents": {
    "defaults": {
      "model": "anthropic/claude-4-5-sonnet-20250929"
    }
  },
  "providers": {
    "claude": {
      "api_key": "your-api-key-here"
    }
  }
}

3. Chat:

golem chat

Or one-shot:

golem chat "Analyze the current directory structure"

4. Go server mode (Telegram + Gateway + Cron):

golem run

---

CLI Commands

Command	Description
golem init	Initialize configuration and workspace
golem chat	Start interactive TUI chat
golem run	Start server mode (Channels + Gateway + Cron)
golem status	Show system status (providers, channels, cron, skills)
golem channels list	List all configured channels
golem channels status	Show detailed channel status
golem cron list	List all scheduled jobs
golem cron add -n <name> -m <msg> [--every <sec> | --cron <expr> | --at <ts>]	Add a scheduled job
golem cron remove <id>	Remove a scheduled job
golem cron enable <id>	Enable a scheduled job
golem cron disable <id>	Disable a scheduled job
golem skills list	List installed skills
golem skills install <repo>	Install a skill from GitHub
golem skills remove <name>	Remove an installed skill
golem skills show <name>	Show skill content
golem skills search [keyword]	Search available remote skills

---

Cron Scheduling

Jobs persist across restarts and can be managed via the CLI or by the agent itself using the manage_cron tool.

Schedule Types

• --every <seconds> — Repeat at a fixed interval (e.g., --every 3600 for hourly).
• --cron <expr> — Standard 5-field cron expression (e.g., --cron "0 9 * * *" for daily at 9 AM).
• --at <timestamp> — One-shot at an RFC3339 timestamp (auto-deleted after run).

Examples

# Hourly system check
golem cron add -n "hourly-check" -m "Check system status and report" --every 3600

# Daily morning briefing
golem cron add -n "morning-brief" -m "Give me a morning briefing" --cron "0 9 * * *"

# One-time reminder
golem cron add -n "meeting" -m "Remind me about the team meeting" --at "2026-02-14T09:00:00Z"

---

Skills System

Skills are markdown instruction packs loaded into the agent's system prompt automatically. Golem discovers skills from three sources with deterministic precedence:

• workspace/skills (highest)
• ~/.golem/skills
• builtin skills directory (default ~/.golem/builtin-skills, overridable via GOLEM_BUILTIN_SKILLS_DIR)

golem init automatically seeds a default builtin skill set into ~/.golem/builtin-skills.

Skill File Format

Each skill lives in workspace/skills/<name>/ with a SKILL.md file:

---
name: weather
description: "Query weather information"
---

# Weather Skill
(Skill instructions for the agent)

Install from GitHub

golem skills install owner/repo

Downloads the SKILL.md from the repository's main branch.

Search Remote Skills

golem skills search
golem skills search weather

By default, the command reads a remote skills.json index. You can override the index URL with GOLEM_SKILLS_INDEX_URL.

---

Configuration

Located at ~/.golem/config.json:

{
  "agents": {
    "defaults": {
      "workspace_mode": "default",
      "model": "anthropic/claude-4-5-sonnet-20250929",
      "max_tokens": 8192,
      "temperature": 0.7
    }
  },
  "channels": {
    "telegram": {
      "enabled": false,
      "token": "YOUR_TELEGRAM_BOT_TOKEN",
      "allow_from": ["YOUR_TELEGRAM_USER_ID"]
    },
    "whatsapp": {
      "enabled": false,
      "bridge_url": "ws://127.0.0.1:8080/ws",
      "allow_from": []
    },
    "feishu": {
      "enabled": false,
      "app_id": "",
      "app_secret": "",
      "encrypt_key": "",
      "verification_token": "",
      "allow_from": []
    },
    "discord": {
      "enabled": false,
      "token": "",
      "allow_from": []
    },
    "slack": {
      "enabled": false,
      "bot_token": "",
      "app_token": "",
      "allow_from": []
    },
    "qq": {
      "enabled": false,
      "app_id": "",
      "app_secret": "",
      "allow_from": []
    },
    "dingtalk": {
      "enabled": false,
      "client_id": "",
      "client_secret": "",
      "allow_from": []
    },
    "maixcam": {
      "enabled": false,
      "host": "0.0.0.0",
      "port": 9000,
      "allow_from": []
    }
  },
  "providers": {
    "openai": { "api_key": "sk-..." },
    "claude": { "api_key": "sk-ant-..." },
    "ollama": { "base_url": "http://localhost:11434" }
  },
  "tools": {
    "exec": {
      "timeout": 60,
      "restrict_to_workspace": true
    },
    "web": {
      "search": {
        "api_key": "YOUR_BRAVE_SEARCH_API_KEY (optional)",
        "max_results": 5
      }
    }
  },
  "gateway": {
    "host": "0.0.0.0",
    "port": 18790,
    "token": "YOUR_GATEWAY_BEARER_TOKEN"
  },
  "log": {
    "level": "info",
    "file": ""
  }
}

---

Gateway API

Available when golem run is active.

Endpoint	Method	Description
/health	GET	Health check
/version	GET	Version info
/chat	POST	Send a message to the agent

POST /chat example:

{
  "message": "Summarize the latest logs",
  "session_id": "ops-room",
  "sender_id": "api-client"
}

If gateway.token is set, include Authorization: Bearer <token>.

---

Development

Local Quality Checks

Run before pushing:

go test ./...
go test -race ./...
go vet ./...

Branch and PR Workflow

1. Create a focused feature branch: feature/<phase>-<topic>
2. Keep PRs small and aligned with one phase/task
3. Merge to main only after CI is green

---

License

MIT