What it does

ZeroClaw is an agent runtime written in Rust and shipped as a single binary that you configure and run on your own machine. It talks to model providers such as Anthropic, OpenAI and Ollama (plus about twenty others and any OpenAI-compatible endpoint), receives messages through 30+ channels including Discord, Telegram, Matrix, email, voice and webhooks, and acts through tools such as shell, browser, HTTP, hardware and custom MCP servers. Setup lives in one TOML file. Default autonomy is supervised: medium-risk operations need approval and high-risk ones are blocked. This is backed by workspace boundaries, command policy, OS-level sandboxes (Landlock, Bubblewrap, Seatbelt, Docker) and cryptographic receipts for tool actions. A YOLO mode removes the safety gates for trusted development machines. Other pieces include an HTTP and WebSocket gateway with a web dashboard for chat, memory browsing, config editing, cron management and tool inspection, an SOP engine for event-triggered procedures with approval gates, editor integration through the Agent Client Protocol, and GPIO, I2C, SPI and USB support on boards such as Raspberry Pi, STM32, Arduino and ESP32. It is dual-licensed under MIT or Apache 2.0.

Capabilities

Single Rust binary configured with one TOML file
Works with Anthropic, OpenAI, Ollama and about 20 other providers, with fallback chains
30+ channels such as Discord, Telegram, Matrix and email feeding one agent
Supervised autonomy with sandboxing and approval gates
Web dashboard and HTTP and WebSocket gateway
Hardware support for Raspberry Pi, STM32, Arduino and ESP32

When to use it

Running a personal assistant on your own machine that answers across several chat channels
Pointing an agent at hosted or local models while keeping keys and data in your own workspace
Building event-triggered agent procedures that pause for approval
Letting an agent drive hardware boards