ZeroClaw Review

ZeroClaw describes itself as an agent runtime, a single Rust binary that you configure and run. It talks to LLM providers (Anthropic, OpenAI, Ollama, and about 20 others), reaches the world through 30-plus channels including Discord, Telegram, Matrix, email, voice, webhooks, and your own CLI, and acts through tools. The README lists its main traits: multi-channel (one agent answering you across every channel you configure, all delivered to the same agent loop), provider-agnostic with pluggable providers and fallback chains and routing, security-first with escape hatches (default autonomy is supervised, where medium-risk operations need approval and high-risk ones are blocked; workspace boundaries; command policy; sandboxes via Landlock, Bubblewrap, Seatbelt, or Docker; and cryptographic tool receipts on every action), and hardware-capable. A YOLO mode skips safety gates for development boxes. The README also points to a Philosophy page for the four opinions that shape the design.

On Unix run the install script with curl and sh, then zeroclaw quickstart to pick a provider and write a working config. On Windows the README recommends a Rust-free prebuilt PowerShell path that installs the release, updates PATH, and runs quickstart. After that, zeroclaw agent -a <alias> starts an interactive chat, and zeroclaw service install and service start register it as a systemd, launchctl, or Windows service. Platform notes exist for Linux, macOS, Windows, FreeBSD, NixOS, and Docker. We did not install it.

The primary interface is whichever channel you connect, plus a CLI. Because it ships as a single binary with a quickstart, setup is more streamlined than frameworks that need several services. We did not run it.

• Single Rust binary
• 30-plus channels and about 20 model providers
• Fallback chains and routing between providers
• Supervised autonomy by default with approval for risky operations
• OS-level sandboxes and cryptographic tool receipts
• Runs as a background service
• Hardware-capable

• Easy deployment compared with multi-service agent stacks
• Safety posture is explicit and layered
• Broad channel and provider coverage in one runtime
• Fast-growing: about 33,000 stars and commits on the day of review

• Young project with a fast-moving surface
• The README is a feature overview, so deep behavior lives in the documentation
• Power comes with risk: YOLO mode removes safety gates
• No managed offering

ZeroClaw is free under Apache-2.0. You pay your model providers and host the runtime yourself.

OpenClaw is a larger, feature-rich personal agent with application-level permissions. ZeroClaw offers a comparable multi-channel assistant as one Rust binary with explicit sandbox layers. NanoClaw takes a third route with a tiny codebase and containers.

• Self-hosters who want an always-on agent across many chat channels
• Security-minded users who want sandboxing
• People who prefer a single binary to a container stack

• Users who want a hosted assistant
• Teams that need a graphical admin console
• Anyone who needs a long-established project