ZeroClaw: The Runtime OS for Agentic Workflows

ZeroClaw is a next-generation AI assistant infrastructure written entirely in Rust. It acts as the runtime operating system for autonomous agents — abstracting models, tools, memory, and execution so you build once and run anywhere. Designed for zero overhead and zero compromise, it ships as a single ~8.8MB binary that runs on $10 hardware with under 5MB RAM (99% less than OpenClaw and similar frameworks).

Perfect for self-hosted, privacy-first deployments on edge devices, servers, or the cloud. Supports 22+ AI providers (OpenAI, Anthropic, Claude, Gemini, Mistral, Ollama, and any OpenAI-compatible endpoint), 30+ messaging channels, and full trait-based swapping of every component.

Features

Blazing Fast & Lean: <10ms cold starts, <5MB RAM typical, sub-millisecond responses
Trait-Driven Pluggability: Swap AI providers, channels, tools, memory backends, runtimes, and tunnels with a config change — no code required
Secure by Default: Pairing authentication, strict sandboxing, filesystem scoping, explicit allowlists, ChaCha20 encryption
Memory System: SQLite hybrid search (vector cosine + BM25 keyword), embeddings, chunking, caching; optional PostgreSQL, Lucid, or Markdown
Channels & Integrations: Telegram, Discord, Slack, WhatsApp (Web/Business API), Matrix, Signal, iMessage, Mattermost, and more
Tools: Shell, file system, git, browser, HTTP, screenshot, cron, hardware peripherals
Gateway & API: Local web dashboard, webhooks, daemon mode, Prometheus/OpenTelemetry observability
Runtimes: Native single binary or Docker sandbox; portable across ARM/x86/RISC-V
Migration Ready: One-command import from OpenClaw (agents, memory, config)
Multi-Language Support: 31 languages with localized READMEs

Performance Comparison

Metric	ZeroClaw	OpenClaw	Typical Python/TS Alternatives
Language	Rust	TypeScript	Python/Go
Binary Size	~8.8 MB	~150 MB+	28 MB+
Runtime RAM	<5 MB	>1 GB	>100 MB
Cold Start (0.8 GHz)	<10 ms	>500 ms	>30 s
Deployment Cost	$10 board	Mac Mini	VPS/SBC

Quick Start

One-Click Installer (Recommended)

curl -fsSL https://zeroclawlabs.ai/install.sh | bash

Homebrew

brew install zeroclaw

From Source (or prebuilt)

git clone https://github.com/zeroclaw-labs/zeroclaw.git
cd zeroclaw
./bootstrap.sh --prefer-prebuilt

First Run

# Onboard with your API key
zeroclaw onboard --api-key sk-... --provider openrouter --model anthropic/claude-sonnet-4-6

# Chat in terminal
zeroclaw agent

# Or start web gateway
zeroclaw gateway

Full CLI commands: zeroclaw status, zeroclaw daemon, zeroclaw channel bind-telegram, zeroclaw migrate openclaw.

Architecture

Everything is a Rust trait: Provider, Channel, Tool, Memory, Runtime, Tunnel, Security. Configure once in ~/.zeroclaw/config.toml and extend with custom implementations. No external runtime dependencies (no Node.js/Python). Supports AIEOS standardized identity format and ClawHub-compatible skills.

Use Cases

Personal Assistant: Always-on Telegram/Discord/WhatsApp bot with persistent memory
Task Automation: Code generation, data analysis, document processing, research agents
Enterprise Bots: Self-hosted customer support or internal tools with strict scoping
Edge AI: Run on Raspberry Pi, low-power SBCs, or air-gapped hardware
Development Workflows: Proactive tool use (git, browser, shell) before responding

Security & Privacy

Binds to localhost by default
Pairing-based auth (no exposed API keys)
Deny-by-default channel/tool allowlists
Workspace isolation and encrypted secrets
Official security policy and checklist in repo docs

Why Choose ZeroClaw?

No cloud subscriptions. No data leaving your machine. Full portability. Built by students and contributors from Harvard, MIT, and Sundai.Club communities. Dual-licensed (MIT + Apache 2.0). Over 26k GitHub stars and growing.

Official repository (single source of truth): github.com/zeroclaw-labs/zeroclaw. Official site: zeroclawlabs.ai. Docs, roadmap, and contributing guidelines included.