🚀 AI-Powered Assistant Infrastructure in Rust

Zero overhead. Zero compromise.

ZeroClaw is a next-generation AI assistant framework written in Rust. Ultra-lightweight, blazing fast, memory-safe, built for production.

Deploy anywhere. Swap anything. Runs on $10 hardware with <5MB RAM — trait-driven architecture, secure by default.

99% less memory than OpenClaw · 98% cheaper than a Mac mini

Visit GitHub Star on GitHub

<10msStartup time
3.4MBBinary size
22+AI providers

Learn

Watch intro and tutorial videos.

Support ZeroClaw

Star the repo to show your support and stay in the loop. One click on GitHub.

Star on GitHub

Why choose ZeroClaw?

Lightweight infrastructure designed for modern AI applications.

Blazing fastCold startup <10ms, instant response.
Ultra lightweightBinary only 3.4MB, minimal resource usage.
Pluggable architectureFlexible component extension based on Rust trait system.
Memory safeGuaranteed by Rust, no data races, no memory leaks.
Multi-platform22+ AI providers, Telegram, Discord, Slack and more.
Zero overheadZero runtime overhead, uncompromised performance.

Teams also get: Secure by design (pairing, sandboxing, allowlists) · Fully swappable traits · No lock-in (OpenAI-compatible + custom endpoints).

Benchmark snapshot

Normalized for 0.8GHz edge hardware. ZeroClaw: Rust, <5MB RAM, <10ms startup, ~3.4MB binary.

	OpenClaw	NanoBot	PicoClaw	ZeroClaw
Language	TypeScript	Python	Go	Rust
RAM	>1GB	>100MB	<10MB	<5MB
Startup (0.8GHz)	>500s	>30s	<1s	<10ms
Binary size	~28MB	N/A	~8MB	~3.4MB
Cost	Mac Mini $599	SBC ~$50	Board $10	Any $10+

Reproducible local measurement — run cargo build --release and /usr/bin/time -l target/release/zeroclaw status on your machine.

Compare ZeroClaw vs OpenClaw, NanoBot, PicoClaw — detailed comparisons and FAQs.

Quick start

One-click bootstrap or build from source. Then onboard and chat.

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

# Or: cargo build --release --locked && cargo install --path . --force --locked
zeroclaw onboard --api-key sk-... --provider openrouter
zeroclaw agent -m "Hello, ZeroClaw!"
zeroclaw daemon   # gateway + channels (Telegram, Discord, Slack…)

Full quick start · One-click bootstrap · Prerequisites

Flexible architecture design

Pluggable architecture based on Rust trait system.

Core layer — Trait-based architecture · Memory safety guarantee · Async runtime.
AI adapter layer — Unified API abstraction · 22+ AI providers · Flexible switching.
Channel layer — Multi-platform support · Message routing · Event handling.

Subsystems at a glance

Subsystem	Trait	Ships with
AI models	`Provider`	28+ built-ins, custom endpoints
Channels	`Channel`	CLI, Telegram, Discord, Slack, Matrix, Signal, WhatsApp, Email, Webhook, …
Memory	`Memory`	SQLite, PostgreSQL, Lucid, Markdown, none
Tools	`Tool`	shell, file, memory, cron, git, browser, composio, hardware, …
Runtime	`RuntimeAdapter`	Native, Docker (sandboxed)
Security	`SecurityPolicy`	Pairing, sandbox, allowlists, encrypted secrets
Integrations	Registry	70+ across 9 categories

Memory is a full-stack search engine: SQLite vector + FTS5 keyword, hybrid merge, no Pinecone or Elasticsearch. Agent recalls, saves, and manages memory via tools.

Full architecture table and diagram in the repo.

Broad support

Seamless integration with mainstream AI providers and communication platforms.

AI providers

OpenRouter Anthropic OpenAI Ollama Gemini Venice Vercel Cloudflare Moonshot Kimi Code Synthetic OpenCode Z.ai GLM MiniMax Bedrock Qianfan Qwen Groq Mistral xAI DeepSeek Together Fireworks Perplexity Cohere Copilot LM Studio NVIDIA

Channels

CLI Telegram Discord Slack Matrix Signal WhatsApp Email Webhook

Security

Security at every layer. ZeroClaw passes the community security checklist.

Gateway not publicly exposed — Binds 127.0.0.1 by default; refuses 0.0.0.0 without tunnel or explicit allow.
Pairing required — 6-digit one-time code; exchange via POST /pair for bearer token; all webhook requests require Authorization.
Filesystem scoped — workspace_only by default; 14 system dirs + 4 sensitive dotfiles blocked; symlink escape detection.
Access via tunnel only — Gateway refuses public bind without active tunnel (Tailscale, Cloudflare, ngrok, custom).

Security checklist and channel allowlists in the repo.

Use cases

IoT & robotsDrones, rovers, fleet dispatch; companion PC or gateway; local or cloud models.

Legal & complianceLegal aid bots, court clerk assistant, RAG over corpus, audit trails.

SaaS & APIAPI-first assistant, white-label bot, managed run-your-own, multi-tenant.

Ops & SREIncident runbook agent, Slack/Teams, approved scripts, audit.

Education & makerClassroom lab in a box, RPi + Arduino; chat to control hardware.

Edge & air-gapLocal models (Ollama, LM Studio); offline queue; burners and field devices.

Tutorials & docs

Quick start Clone, build, onboard, first message. One-click bootstrap Bootstrap script and toolchain. Getting started First-time setup and install. Hardware & peripherals RPi, Nucleo, Arduino, GPIO. Network deployment Deploy on a server or device. Docs TOC Full documentation index.

Featured ideas

Use case ideas, features, and hackathon bounties. Pick one and build. Search all ideas →

View all ideas & search

Contribution guidelines

ZeroClaw is open source. We welcome contributions that follow our architecture and quality bar.

Setup: Clone, enable pre-push hook (.githooks), cargo build, cargo test --locked.
Quality: Run ./scripts/ci/rust_quality_gate.sh and cargo fmt before PR.
Extend by trait: New Provider → src/providers/; new Channel → src/channels/; new Tool → src/tools/.
Docs: Update reference docs when behavior or config changes; keep SUMMARY and README in sync.

Full contributing guide

Get started with ZeroClaw

Next-generation AI assistant infrastructure built with Rust, designed for production.

Visit GitHub repository