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
Unix (recommended)
curl -fsSL https://raw.githubusercontent.com/zeroclaw-labs/zeroclaw/master/install.sh | sh
"${CARGO_HOME:-$HOME/.cargo}/bin/zeroclaw" quickstart
Windows (recommended)
Use the Rust-free prebuilt PowerShell path in the Windows setup guide. It installs the current release, updates PATH, and runs Quickstart.
For the guided Unix installer, source builds, app and feature selection, and PATH behavior, compare the installation paths.
Run ./install.sh --help for the complete Unix installer flag reference.
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:
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 background
Full 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 Peripheral trait. 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.
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-5.4
The 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_key on the provider entry.
Claude Max setup-token auth stays on the canonical Anthropic slot: run claude setup-token, choose setup_token in Quickstart, and paste the generated token into the API key/token prompt.
Only set api_key / uri on [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. Check zeroclaw auth status before changing provider config.
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.
ZeroClaw 是一个个人 AI 助手运行时：它是一个单独的 Rust 二进制程序，由你自己配置和运行，可以与大语言模型服务商通信，监听数十个渠道，并通过 shell、浏览器和 HTTP 等工具执行操作。