Fast, small, and fully autonomous AI assistant infrastructure with <5MB RAM footprint, <10ms cold start, 20+ messaging channels, and 15+ LLM providers, deployable to edge devices.
ZeroClaw is an AI assistant runtime infrastructure built in Rust (95%), featuring a Trait-Driven architecture with fully pluggable Provider/Channel/Tool/Memory/Tunnel/Observer/Runtime/Identity/Heartbeat/Skills components.
Core Features#
Lean Runtime: <5MB RAM for CLI/state operations, <10ms cold start (0.8GHz core), ~8.8MB release binary.
Secure-by-Design: Gateway binds to 127.0.0.1 by default (0.0.0.0 blocked), mandatory pairing (6-digit one-time code) for Bearer Token, filesystem isolation (workspace_only = true), blacklist directories, symlink escape detection, null byte injection blocking, Channel whitelist deny-by-default.
Pluggable Providers: Built-in OpenAI, Anthropic, OpenRouter, Google AI, Mistral, Groq, xAI, DeepSeek, Cohere, Perplexity, Ollama (local/remote), llama.cpp, vLLM, Osaurus; custom endpoints via custom:https://your-api.com (OpenAI-compatible) or anthropic-custom:https://your-api.com.
Multi-Channel Support: CLI, Telegram, Discord, Slack, Mattermost, iMessage, Matrix, Signal, WhatsApp (Web/Business Cloud API), Linq, Email, IRC, Lark, DingTalk, QQ, Nostr, Webhook, and 20+ more.
Memory System: SQLite hybrid search (vector + FTS5 BM25), PostgreSQL backend, Lucid bridge, Markdown files, none backend; proprietary Embedding + FTS5 + hybrid merge, no external vector DB dependency.
Built-in Tools: shell/file/memory, cron/schedule, git, pushover, browser, http_request, screenshot/image_info, composio (optional), delegate, hardware, etc.
Runtime Modes: native (local direct execution), docker (sandboxed container with Alpine image, network isolation, read-only rootfs, resource limits); WASM/edge runtime planned.
Skills Ecosystem: TOML manifest + SKILL.md directives; community open-skills sync requires explicit enablement; static security audit during installation.
Use Cases#
- Edge/embedded deployment: $10-class SBCs, Raspberry Pi, small cloud instances
- Local-first AI assistant: data stays on-premise, fully private
- Code generation/refactoring/review: with shell/git/file tools
- Multi-channel bot gateway: unified Agent backend
- Scheduled tasks/heartbeat: HEARTBEAT.md + cron management
- Data-sensitive workflows: local processing
Installation#
- Homebrew:
brew install zeroclaw - Bootstrap script:
./bootstrap.sh(supports --docker, --onboard, --prefer-prebuilt, etc.) - Prebuilt binaries: multi-arch Linux/macOS/Windows
- From source:
cargo build --release --locked
Security Notice#
Official sources warn of impersonating repos/sites (e.g., openagen/zeroclaw, zeroclaw.org/.net). Only github.com/zeroclaw-labs/zeroclaw is the official source.