OdinClaw

Small enough to run on your coffee maker.

406 KB binary · 2.7s build · 5.4 MB Docker image

AI assistant runtime written in Odin, inspired by NullClaw (Zig).

📖 Detailed Guide: See AGENTS.md for comprehensive operational documentation.

Features

Tiny: 406 KB binary, 5.4 MB Docker image, 2.7s build on Apple Silicon
Multi-Provider Support: OpenAI, Anthropic, xAI, Ollama, and 30+ OpenAI-compatible providers
10 Built-in Tools: shell, file_read, file_write, file_edit, file_append, git, http_request, memory_store, memory_recall, memory_forget
Multiple Channels: CLI, Telegram, Slack, Discord
LMDB Backend: Fast local memory storage
Gateway Server: HTTP server with CORS, pairing, rate limiting
Security: ChaCha20-Poly1305 encryption, audit logging, path sandboxing

Installation

# Compile C helper for libcurl (one-time)
cc -c src/curl_helpers.c -o src/curl_helpers.o

# Build (optimized for speed)
odin build src -out:odin-claw -o:aggressive

Configuration

Create ~/.odin-claw/config.json:

{
  "providers": {
    "default_provider": "openai",
    "default_model": "gpt-4o",
    "openai_api_key": "sk-..."
  },
  "tools": {
    "workspace_path": "/tmp"
  },
  "channels": {
    "telegram_api_key": "..."
  }
}

Or use environment variables:

OPENAI_API_KEY
ANTHROPIC_API_KEY
XAI_API_KEY
NULLCLAW_PROVIDERS_DEFAULT_PROVIDER

Commands

Command	Description
`odin-claw agent`	Start the AI agent loop
`odin-claw gateway`	Start HTTP gateway server
`odin-claw channel`	Manage messaging channels
`odin-claw cron`	Manage scheduled tasks
`odin-claw status`	Show system status
`odin-claw doctor`	Run diagnostics
`odin-claw onboard`	Initial setup wizard
`odin-claw skills`	List/manage skills
`odin-claw hardware`	Hardware management
`odin-claw migrate`	Data migration
`odin-claw models`	List available models
`odin-claw daemon`	Run as background daemon
`odin-claw service`	Run as system service
`odin-claw help`	Show help

Providers

OpenAI: gpt-4o, gpt-4-turbo, gpt-3.5-turbo
Anthropic: claude-sonnet-4, claude-opus-4
xAI: grok-2, grok-beta
Ollama: llama3, mistral, codellama (local)
OpenAI-Compatible: Groq, DeepSeek, LM Studio, Vercel, Mistral, Cohere, Together, Perplexity, NVIDIA NIM, Venice, Zen/OpenCode

Tools

All tools support sandboxed execution with path validation:

shell - Execute shell commands
file_read - Read files
file_write - Write files
file_edit - Edit files (replace text)
file_append - Append to files
git - Git operations (status, commit, push, pull)
http_request - Make HTTP requests
memory_store - Store in memory DB
memory_recall - Search memory
memory_forget - Delete from memory

Testing

Run comprehensive tests (25 total in separate test files):

odin test src -all-packages

Tests are organized in separate files (*_test.odin) to keep the runtime binary lean and fast.

Development

Project Structure

src/
├── main.odin          # CLI entry point
├── agent.odin         # AI agent orchestration
├── providers.odin     # LLM provider integrations
├── tools.odin         # Tool definitions
├── memory.odin        # In-memory storage
├── lmdb.odin          # LMDB backend
├── channels.odin      # Messaging channels
├── gateway.odin       # HTTP server
├── config.odin        # Configuration management
├── runtime.odin       # Process execution
├── security.odin      # Security utilities
├── peripherals.odin   # Hardware interfaces
├── curl_helpers.c     # C libcurl bindings
└── *_test.odin        # Test files (separate from runtime)

Building

# Development build
odin build src -out:odin-claw

# Optimized release build
odin build src -out:odin-claw -o:aggressive

License

MIT