ZeroClaw — Personal AI assistant runtime
A single Rust binary that runs your own AI assistant across 30+ channels, ~20 model providers, and every major OS, with your data, keys, and machine.
- Source repo
- zeroclaw-labs/zeroclaw
- Stars
- ★ 33k
- Last updated
- today
- License
- Apache-2.0
- Primary language
- Rust
- FA score
- 60/100 · Some gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformOpenAI API · Claude API
- Cost
- Free software; you pay for model usage
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- Individual developers wanting a personal assistant that works across Discord, Telegram, and CLI while retaining full ownership of their data.
- Not a fit if
- Users who want a managed service and no self-hosting upkeep
- People unwilling to edit TOML config or manage doc submodules
- Source review
- 60/100 · Some gaps
What does this agent do, and when should you use it?
ZeroClaw is a self-hosted AI assistant runtime written in Rust, distributed as a single binary. It ingests messages from 30+ channels (Discord, Telegram, Matrix, email, webhooks, CLI) and connects to model providers like Anthropic, OpenAI, Ollama, and ~20 others. Security-first, it defaults to supervised autonomy with sandboxing (Landlock/Bubblewrap/Seatbelt/Docker) and cryptographic tool receipts. It also provides an HTTP/WebSocket gateway and web dashboard for chat, memory browsing, and config editing. ZeroClaw runs on Linux, macOS, Windows, FreeBSD, NixOS, and Docker, and supports hardware interfaces like GPIO/I2C/SPI/USB.
ZeroClaw lets you define multiple agents, channels, model providers, and tools through a single TOML config file. zeroclaw quickstart guides you through provider and agent setup. zeroclaw agent -a <alias> starts interactive chat, while zeroclaw service install registers the agent as a system service. Its event-driven SOP engine reacts to MQTT, webhook, cron, or peripheral events with approval gates and resumable runs. Tools include shell, browser, HTTP, hardware, and custom MCP servers. It integrates with IDEs via the Agent Client Protocol (JSON-RPC 2.0 over stdio).
- Individual developers wanting a personal assistant that works across Discord, Telegram, and CLI while retaining full ownership of their data.
- Teams needing self-hosted automation triggered by webhooks or cron with human approval gates for medium-risk tasks.
- Hardware enthusiasts who want to control Raspberry Pi GPIO, I2C, SPI, or USB peripherals.
- Organizations that need to securely access multiple LLM providers with fallback chains for reliability.
- Developers who want an AI assistant directly inside their IDE via ACP.
How do you install or deploy this agent?
On Unix, run curl -fsSL https://raw.githubusercontent.com/zeroclaw-labs/zeroclaw/master/install.sh | sh, then use ${CARGO_HOME:-$HOME/.cargo}/bin/zeroclaw for Quickstart. Windows users should follow the prebuilt PowerShell path guide. You'll need a valid API key for at least one model provider (e.g., OpenAI, Anthropic, or local Ollama).
How do you use this agent?
Run zeroclaw quickstart to set up your config. Edit ~/.zeroclaw/config.toml to adjust providers, channels, and security settings. Install as a service with zeroclaw service install and start it with zeroclaw service start. To chat, use zeroclaw agent -a <alias>. For OpenAI Codex subscription auth, you can import existing credentials via zeroclaw auth login --model-provider openai-codex --import ~/.codex/auth.json, or use a Claude Max setup token with claude setup-token.
What are this agent's strengths and limitations?
- Bundles 30+ channels and ~20 model providers in one binary.
- Security-first with sandboxing, approval gates, and cryptographic tool receipts.
- Event-driven SOP engine for complex automation with resumable runs.
- Supports a wide range of hardware and platforms, from Raspberry Pi to cloud servers.
- Requires self-hosting and operational responsibility.
- Learning curve for configuring multiple TOML sections and submodules.
- Dependence on external model provider APIs and availability.
- Advanced features like hardware support require specific setup and may not work on all platforms.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| ZeroClaw — Personal AI assistant runtime This agent | 60 · Some gaps | CLIFree + model costs | ★ 33k | today | Rust | OpenAI API · Claude API |
| Moltis Personal Agent Server | 53 · Major gaps | Self-hosted serviceFree + model costs | ★ 2.9k | today | Rust | Codex · Claude Code |
| JoySafeter | 59 · Major gaps | Self-hosted serviceFree + model costs | ★ 314 | 6mo ago | Python | — |
| Jelly: Local Multi-Bot Agent Workspace | 52 · Major gaps | — | ★ 60 | 6d ago | TypeScript | ChatGPT · OpenAI API |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows default supervised mode, workspace boundaries, command allowlisting, forbidden path list, rate limiting, but lacks implementation details or test evidence. Deductions: least privilege not explicitly stated, user confirmation mechanism not detailed, data flow transparency insufficient, sensitive data handling (e.g., secret store) mentioned but not verified, dependency security has pinned versions but no vulnerability scan evidence, external effects have sandbox but effectiveness not proven, rollback only mentioned for update pipeline, source attribution unverified.
Self-consistency: README, Cargo.toml, SECURITY.md are consistent, but no runtime evidence. Dependency availability: versions pinned, but availability not verified. Failure messages: health checks and error handling present, but no specific failure message examples. Deductions: no actual runtime tests, so self-consistency and dependency availability cannot be confirmed.
Target audience clear (personal AI assistant), scenarios diverse (multi-channel, multi-provider, hardware support), capability boundaries described (tools, channels, providers), trigger precision has configuration docs, environment fit has platform-specific docs. Deductions: no runtime verification, capability boundaries may be incomplete.
Information architecture clear (README, docs, architecture diagram), install notes detailed, naming stable (zeroclaw), examples and FAQ present, known limitations not explicitly listed, license clear (MIT OR Apache-2.0), version number exists but no changelog, maintenance responsibility mentioned (community). Deductions: known limitations not explicit, version changelog missing.
Output usability: CLI, gateway, dashboard present, but no actual output examples. Marginal value: feature-rich, but no comparison with alternatives. Cost-benefit: performance optimizations (binary size) mentioned, but no actual data. Deductions: no runtime evidence, so output usability and cost-benefit cannot be assessed.
Claim traceability: README claims have doc links, but no specific code references. Cross-source corroboration: CI and test scripts present, but no independent verification. Fact-inference separation: docs distinguish facts and inferences, but not explicitly labeled. Deductions: cross-source corroboration insufficient, fact-inference separation not explicit.
- No runtime tests provided; all assessments are based on static source, confidence is low.
- Security mechanisms (e.g., sandbox, command allowlisting) are described but lack implementation details or test evidence; rely with caution.
- Dependency versions are pinned, but no vulnerability scan report is provided; check dependency security yourself.
- Known limitations are not explicitly listed, potentially hiding issues.
- Version changelog is missing, making it hard to track feature changes and fixes.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
ZeroClaw v0.8.6 ZeroClaw v0.8.6 is a plugin, identity, and runtime-ownership release spanning 573 commits from 74 contributors. The seven 64-bit CLI release archives now carry the WebAssembly plugin host, and the plugin registry offers channel and tool packages built for this release. It also adds authenticated principals with OIDC across RPC and the gateway, moves live configuration, events, and agent lifecycle under the daemon, and adds an opt-in minimal built-in tool set while full stays the default. Several security tightenings change defaults; review [Breaking Changes](#breakin …View on GitHub ↗ -
Release notes excerpt
ZeroClaw v0.8.5 ZeroClaw v0.8.5 is a security, connectivity, and operator-experience release spanning 454 commits from 73 contributors. It introduces ZeroRelay and ZeroRouter, expands live chat and provider capabilities, hardens plugin, sandbox, webhook, credential, and file boundaries, and makes cross-platform builds and coordinated crate publication more reproducible. This release also removes or narrows several deprecated surfaces; review [Breaking Changes](#breaking-changes) before upgrading plugins, skills, legacy node integrations, WATI deployments, TodoWrite configuration, or …View on GitHub ↗ -
Release notes excerpt
ZeroClaw v0.8.4 ZeroClaw v0.8.4 is a maintenance and hardening release spanning 262 commits from 49 contributors. It expands the memory and SOP control planes, improves provider and channel reliability, strengthens sandbox and credential boundaries, and makes the desktop and release pipeline more dependable. This release also includes upgrade-sensitive changes for skill sources, Nextcloud Talk, and generic webhook authentication; review [Breaking Changes](#breaking-changes) before upgrading those integrations. Highlights …View on GitHub ↗
FAQ
How does ZeroClaw ensure security?
Is ZeroClaw compatible with OpenAI Codex subscriptions?
zeroclaw auth login --model-provider openai-codex --import ~/.codex/auth.json.