Overview

ZeroClaw is an agent runtime shipped as a single Rust binary, dual-licensed MIT or Apache-2.0; its README says it is built and maintained by the community, and ZeroClaw Labs holds the name. It talks to Anthropic, OpenAI, Ollama, Bedrock, Gemini, Azure, OpenRouter and any OpenAI-compatible endpoint, reaches more than 30 channels and keeps memory in SQLite by default. Out of the box medium-risk actions need approval, high-risk ones are blocked, file tools stay in the workspace and tools run in an OS sandbox it detects; network egress stays open unless you restrict it, and its audit log covers certificate issuance only. Prebuilt binaries cover Linux, macOS, Windows and every Raspberry Pi, the Zero 2 W included. Until October 2026 this site called it a local-only agent from a privacy collective, AGPL or BSL licensed, with egress blocked at iptables: its repository says none of that.

Our take

Editorial opinion, not a test result.

Best for

Small or low-power hosts, down to a Raspberry Pi Zero 2 W, that benefit from a single Rust binary
Teams that want to swap model providers freely, a local Ollama included
Hardware projects: it drives GPIO, I2C, SPI and USB on Raspberry Pi, STM32, Arduino and ESP32 boards

Not for

Teams that need the largest plugin ecosystem (OpenClaw)
Python-first teams (Hermes Agent, Nanobot)
Setups that need command execution audited: its audit log covers certificate issuance only

Security posture

Sandbox by default	yes
Auth on dashboard	no
Threat model documented	yes
Multi-LLM support	yes
CVEs on NVD (total)	1
CVEs in the last 30 days	0

On by default: tools run inside the best OS sandbox the host offers (Landlock, Bubblewrap, Firejail or Docker on Linux, Seatbelt on macOS, AppContainer on Windows, experimental), under the default Supervised autonomy and a workspace-only rule. Sandboxed tools keep full network egress unless you pick a backend that blocks it or restrict each tool's domains.