ZeroClaw is an open-source agent runtime: a single Rust binary you configure and run. Its slogan is blunt about the point of it — you own the agent, you own the data, you own the machine it runs on.

Overview

ZeroClaw sits in the same family as OpenClaw — a personal, self-hosted AI assistant rather than a hosted product — but takes a different engineering approach. Where OpenClaw is a TypeScript project running on Node.js, ZeroClaw is written in Rust and compiles to one small binary. The project describes itself as "fast, small, and fully autonomous AI personal assistant infrastructure, any OS, any platform — deploy anywhere, swap anything."

The runtime talks to LLM providers (Anthropic, OpenAI, Ollama, and around twenty others), reaches the world through 30+ channels (Discord, Telegram, Matrix, email, voice, webhooks, or your own CLI), and acts through tools (shell, browser, HTTP, hardware, and custom MCP servers). Everything runs on your machine, with your keys, in your workspace.

The repository was created in February 2026 and is dual-licensed MIT OR Apache-2.0.

Key Features

One Rust Binary: No language runtime to install alongside it. A kernel-only build is roughly 6.6 MB.
Swap Anything: Providers, channels and tools are all pluggable. Around 20 LLM providers are supported, including local inference through Ollama.
30+ Channels: Discord, Telegram, Matrix, email, voice, webhooks, and a plain CLI.
Tools: Shell execution, browser control, HTTP requests, hardware access, and any custom MCP server.
Feature Flags: Build exactly the runtime you need — --preset minimal for a kernel-only build, or a custom feature set such as --features agent-runtime,channel-discord.
Skills Registry: ZeroClaw Labs maintains an official skill registry at zeroclaw-labs/zeroclaw-skills, with community-contributed skills, tools and workflows.
Optional Gateway and TUI: The gateway can be compiled in or out, and zerocode ships as an optional terminal UI app.

How It Works

ZeroClaw is a runtime, not a chat app. You configure a workspace, point it at a model provider, wire up the channels you want to reach it on, and grant it the tools it may use. From there it runs on hardware you control.

Technical Architecture:

Language: Rust (edition 2024).
Licence: MIT OR Apache-2.0.
Providers: Anthropic, OpenAI, Ollama, and roughly 20 others.
Channels: 30+, including Discord, Telegram, Matrix, email, voice and webhooks.
Tools: Shell, browser, HTTP, hardware, and custom MCP servers.
Build: Prebuilt binaries, or a source build with selectable feature flags.

Use Cases

Privacy-First Personal Assistant

Run an always-on assistant on your own hardware, with your own API keys, without a vendor intermediating your conversations.
Point it at a local model through Ollama so nothing leaves the machine at all.

Small and Embedded Deployments

The minimal build is small enough to sit comfortably on constrained hardware — the main practical argument for a Rust runtime over Node-based alternatives.
Compile out the gateway and the TUI when you only need a headless runtime.

Developer Infrastructure

Give the agent shell and HTTP tools and use it to run tests, watch deployments, and report back over Discord or Telegram.
Extend it with MCP servers that expose your own internal systems.

Chat-Ops

Attach the agent to a Matrix room, a Discord guild or an email inbox and let a team interact with it where they already work.

Getting Started

Install

curl -fsSL https://raw.githubusercontent.com/zeroclaw-labs/zeroclaw/master/install.sh | bash

Or clone and run the installer yourself:

git clone https://github.com/zeroclaw-labs/zeroclaw.git
cd zeroclaw
./install.sh

The installer asks whether you want a prebuilt binary (fast) or a source build (slower, customisable). Either way it finishes by launching zeroclaw quickstart.

Useful Installer Flags

./install.sh --prebuilt              # always prebuilt; don't ask
./install.sh --source                # always build from source
./install.sh --preset minimal        # kernel-only source build (~6.6 MB)
./install.sh --source --features agent-runtime,channel-discord
./install.sh --apps zerocode         # select apps to install; "none" to skip all
./install.sh --without-tui           # skip building zerocode
./install.sh --with-gateway          # force gateway support on
./install.sh --prefix /tmp/zc-test   # install under a custom prefix
./install.sh --dry-run --prebuilt    # preview without installing
./install.sh --list-features         # print available feature flags
./install.sh --uninstall             # remove ZeroClaw

Configure

Run zeroclaw quickstart (the installer does this for you) and follow the prompts to choose a model provider, add your API key, and enable the channels you want.

Best Practices

Start minimal. Build only the features you need; every channel you compile in is another way into the agent.
Scope the tools. Shell and browser tools are powerful. Grant them deliberately.
Use a local model first to understand the agent's behaviour before spending on a cloud provider.

Pricing & Access

ZeroClaw: Free and open source (MIT OR Apache-2.0). There is no paid tier.
Self-Hosted: You pay only for your own hardware and electricity.
Inference Costs: If you point ZeroClaw at a cloud provider, you pay that provider directly. Point it at Ollama and inference is free.

Limitations

Terminal-First: Setup is a command-line exercise. The official project ships no consumer mobile app.
Young Project: The repository was created in February 2026. Interfaces and feature flags are still moving.
Local Model Capability: Small local models are noticeably less capable than frontier cloud models on hard reasoning.
Security Surface: An autonomous agent with shell access, browser control and messaging accounts is a serious trust decision. Configure it deliberately.

Community & Support

GitHub: github.com/zeroclaw-labs/zeroclaw
Website: zeroclawlabs.ai
Documentation: docs.zeroclawlabs.ai
Skills Registry: zeroclaw-labs/zeroclaw-skills
Discord: discord.gg/zeroclaw

ZeroClaw