zeroclaw-labs/zeroclaw

What it solves

ZeroClaw provides a self-hosted, privacy-focused agent runtime that allows users to own their AI assistant, their data, and the hardware it runs on. It eliminates dependency on centralized AI platforms by enabling a single binary to manage multiple LLM providers, communication channels, and local system tools.

How it works

It operates as a Rust-based runtime that connects LLM providers (such as Anthropic, OpenAI, and Ollama) to various interaction channels (like Discord, Telegram, and CLI) and actionable tools. The system uses a TOML configuration file to define agents, risk profiles, and provider routing. It includes a security layer with supervised autonomy and OS-level sandboxing to ensure that agent actions are gated and safe.

Who it’s for

It is designed for users who want a highly customizable, private AI agent that can interact with their local machine, hardware peripherals, and multiple messaging apps simultaneously.

Highlights

• Multi-channel support: One agent can be reached via Discord, Telegram, Matrix, email, and more.
• Provider-agnostic: Pluggable support for over 20 LLM providers with fallback chains to ensure reliability.
• Hardware integration: Ability to control GPIO, I2C, SPI, and USB on devices like Raspberry Pi and Arduino.
• Security-first: Features supervised autonomy, tool receipts, and sandboxing via Landlock, Bubblewrap, or Docker.
• SOP engine: Event-triggered Standard Operating Procedures for automated workflows with approval gates.
• ACP integration: IDE and editor support via the Agent Client Protocol (JSON-RPC 2.0).