🦀 ZeroClaw — Personal AI Assistant
You own the agent. You own the data. You own the machine it runs on.
Docs · Philosophy · Quick start · Architecture · Discord
ZeroClaw is an agent runtime — a single Rust binary you configure and run. It talks to LLM providers (Anthropic, OpenAI, Ollama, and ~20 others), reaches the world through 30+ channels (Discord, Telegram, Matrix, email, voice, webhooks, your own CLI), and acts through tools (shell, browser, HTTP, hardware, custom MCP servers). Everything runs on your machine, with your keys, in your workspace.
Read the Philosophy for the four opinions that shape it.
Install
curl -fsSL https://raw.githubusercontent.com/zeroclaw-labs/zeroclaw/master/install.sh | bashOr clone and run:
git clone https://github.com/zeroclaw-labs/zeroclaw.git
cd zeroclaw
./install.shThe piped installer uses a prebuilt binary when one is available and falls back to a source build otherwise. It skips interactive setup and prints zeroclaw quickstart as the next step.
When the platform maps to a supported prebuilt target, running ./install.sh from a clone in an interactive terminal offers prebuilt or source installation; other platforms build from source. The source path also lets you select apps and optional features. For an unconfigured install, the installer then offers CLI or browser-based Quickstart. Use --skip-quickstart when you only want to install.
Working on the docs? The translated documentation catalogues live in a git submodule (
docs/book/po). The Rust build does not need it, but building or syncing the docs does. Clone with it, or add it to an existing clone:git clone --recurse-submodules https://github.com/zeroclaw-labs/zeroclaw.git git submodule update --init docs/book/po # existing clone
Flags:
./install.sh --prebuilt # always prebuilt; don't ask
./install.sh --source # always build from source
./install.sh --preset minimal # kernel-only source preset (~6.6 MB)
./install.sh --minimal # alias for --preset minimal
./install.sh --source --features agent-runtime,channel-discord # custom feature set
./install.sh --apps zerocode # select apps to install; use "none" to skip all
./install.sh --without-tui # skip building zerocode
./install.sh --with-gateway # force gateway support on
./install.sh --without-gateway # force gateway support off
./install.sh --prefix /tmp/zc-test # install under a custom prefix
./install.sh --dry-run --prebuilt # preview without installing
./install.sh --skip-quickstart # install only, run `zeroclaw quickstart` later
./install.sh --list-features # print available feature flags
./install.sh --uninstall # remove ZeroClaw
Platform-specific notes: Linux · macOS · Windows · FreeBSD · NixOS · Docker
Quick start
zeroclaw quickstart # one-shot setup: pick a provider, write a working config
zeroclaw agent -a <alias> # interactive chat using the [agents.<alias>] entry
zeroclaw service install # register as systemd/launchctl/Windows Service
zeroclaw service start # run it always-on in the backgroundFull walkthrough: Quick start — or skip the safety gates with YOLO mode for dev boxes.
What ZeroClaw does
- Multi-channel — one agent answering you across every channel you configure. Inbound messages from Discord, Telegram, Matrix, email, webhooks, CLI — all delivered to the same agent loop.
- Provider-agnostic — model providers are pluggable. Configure Anthropic, OpenAI, local Ollama, or any OpenAI-compatible endpoint. Fallback chains and routing keep the agent running when a provider flakes.
- Security-first, with escape hatches — default autonomy is
supervised: medium-risk ops require approval, high-risk blocked. Workspace boundaries, command policy, OS-level sandboxes (Landlock / Bubblewrap / Seatbelt / Docker), and cryptographic tool receipts on every action. YOLO mode exists for trusted dev environments. - Hardware-capable — GPIO / I2C / SPI / USB on Raspberry Pi, STM32, Arduino, and ESP32 via the
Peripheraltrait. See Hardware. - Gateway + dashboard — HTTP / WebSocket gateway for clients, with a web dashboard for chat, memory browsing, config editing, cron management, and tool inspection.
- SOP engine — event-triggered Standard Operating Procedures (MQTT / webhook / cron / peripheral) with approval gates and resumable runs.
- ACP — IDE / editor integration via Agent Client Protocol (JSON-RPC 2.0 over stdio).
Configuration
One TOML file at ~/.zeroclaw/config.toml. Pointers:
- Provider configuration — the universal
[providers.models.<type>.<alias>]schema - Channels overview — per-channel
[channels.<type>.<alias>]blocks - Security overview — autonomy, sandboxing, tool receipts
- Full config reference — generated from the live schema; every key documented
A V3 config has at minimum four section headers (<type>.<alias> shaped) — a provider entry, an agent that references it, and a risk profile the agent gates against. See Provider Configuration → Minimal working example for the canonical four-section form with inline type/alias commentary.
For standard OpenAI Codex subscription auth, Quickstart can write the provider entry for you:
zeroclaw auth login --model-provider openai-codex --import ~/.codex/auth.json # if already signed in with Codex CLI
zeroclaw quickstart --model-provider openai-codex --model gpt-5.4The provider entry uses the canonical OpenAI shape; the alias below is an example:
[providers.models.openai.coding] # type = openai; alias = coding (you choose)
model = "gpt-5.4"
wire_api = "responses"
requires_openai_auth = true…and point your agent at it with model_provider = "openai.<alias>".
Notes:
- Normal OpenAI Codex subscription auth uses stored auth profiles, not an
api_keyon the provider entry. - Claude Max setup-token auth stays on the canonical Anthropic slot: run
claude setup-token, choosesetup_tokenin Quickstart, and paste the generated token into the API key/token prompt. - Only set
api_key/urion[providers.models.openai.<alias>]when intentionally targeting a custom OpenAI-compatible gateway or endpoint. - If you see
provider streaming failed, falling back to non-streaming chat, ZeroClaw retries the same request in non-streaming mode. Checkzeroclaw auth statusbefore changing provider config.
Architecture
┌──────────────────────────────────────────────────────────────┐
│ channels gateway ACP │
│ (30+ adapters) (REST/WS) (JSON-RPC) │
│ ↓ │
│ ZeroClaw runtime │
│ ┌──────────┬──────────┬──────────┐ │
│ │ agent │ security │ SOP │ │
│ │ loop │ policy │ engine │ │
│ └──────────┴──────────┴──────────┘ │
│ ↓ ↓ ↓ │
│ providers tools memory │
│ (Anthropic, (shell, (SQLite, │
│ OpenAI, browser, embeddings) │
│ Ollama, HTTP, │
│ ~20 more) hardware) │
└──────────────────────────────────────────────────────────────┘
Full detail with Mermaid diagrams: Architecture overview · Request lifecycle · Crates.
Contributing
Start with how to contribute. Larger changes go through the RFC process. Real-time chat lives on Discord (the best way to reach the team); durable work tracking is on GitHub issues.
Good places to start:
- New channel →
crates/zeroclaw-channels/ - New provider →
crates/zeroclaw-providers/ - New tool →
crates/zeroclaw-tools/ - Hardware support →
crates/zeroclaw-hardware/ - Docs →
docs/book/src/
AI-assisted PRs are welcome; see Contribution culture (RFC #5615) for the co-authorship norms.
Security
Do not file public issues for security vulnerabilities. Email security@zeroclaw.dev. See SECURITY.md for the full policy.
Official repository & impersonation notice
This is the only official ZeroClaw repository:
Any other repository, organization, domain, or package claiming to be "ZeroClaw" or implying affiliation with ZeroClaw Labs is unauthorized and not affiliated with this project.
License
Dual-licensed: MIT OR Apache 2.0. You may choose either. Contributors automatically grant rights under both — see CLA. The ZeroClaw name and logo are trademarks of ZeroClaw Labs.
Credits
Built and maintained by the community — original creator @theonlyhennygod; project lead @JordanTheJet. Full maintainer list in Communication.
Thanks to the communities that incubated early work: Harvard University, MIT, Sundai Club, and every contributor pushing it forward.
