Installer zeroclaw avec Homebrew, Nix, scoop, winget

Liens du projet

• Page d'accueil
• Dépôt
• Docs amont
• Page du gestionnaire de paquets

Commandes et alias

• zeroclaw

Historique du projet

ZeroClaw's public repository describes it as fast, small, autonomous personal-assistant infrastructure that can run across operating systems and platforms. The README presents an architecture with channels, gateway, ACP/JSON-RPC, runtime, agent loop, security policy, SOP engine, providers, tools, and memory.

The project is new enough that most reliable history is in its own repository. By June 2026, the project had moved documentation out of the GitHub wiki to docs.zeroclawlabs.ai and the README identified the GitHub repository as the only official ZeroClaw repository, which is useful because the name appears in adjacent and unofficial pages.

Its security docs frame the core problem directly: an agent that can execute shell commands, open URLs, and write files is privileged, so every tool call and channel message is gated by risk/runtime profiles and security policy.

Historique d'adoption

The README shows a large public GitHub footprint by mid-2026, with many stars, forks, releases, discussions, issues, and pull requests. The supplied package data lists Homebrew, Nix, Scoop, and winget packaging, suggesting early cross-platform packaging for people evaluating agent runtimes outside a single language ecosystem.

The adoption story is still early and should be treated as active-project evidence rather than mature ecosystem history. Its relevance is strongest among users testing local AI assistants, multi-channel agent gateways, and Rust-native alternatives to heavier Node/Python assistant stacks.

Modes d'utilisation

Operators use the `zeroclaw` CLI to onboard, configure providers and secrets, run local agents, expose gateway/API modes, and wire channels and tools into the runtime. The repository separates CLI commands for operators from tools that the agent can invoke mid-conversation.

Tools are treated as privileged capabilities, such as shell, HTTP, browser, file, PDF, or hardware access, and produce receipts while passing through security policy. That makes the package closer to an agent runtime than a simple chatbot CLI.

Pourquoi les passionnés de paquets s'y intéressent

ZeroClaw matters as a package because it packages a whole agent runtime into a CLI: local state, provider configuration, secrets, tools, memory, channel adapters, gateway surfaces, and security policy. Installing it is not just adding a command; it is installing an operator entrypoint for autonomous actions.

The project also shows the 2026 packaging pressure around AI tools: users want Homebrew/Nix/Scoop/winget convenience, but maintainers need clear official-source identity, documentation location, and safe defaults for a tool that may execute commands and touch files.

Chronologie

• 2026-06-17: GitHub wiki page says ZeroClaw documentation moved to docs.zeroclawlabs.ai as the single source of truth.
• 2026-06-26: GitHub repository release listing shows v0.8.2 as latest.
• 2026: README identifies the GitHub repository as the only official ZeroClaw repository and documents architecture, contribution areas, and security policy.

Related projects

• OpenClaw and other Claw-family agent projects are adjacent points of comparison for local assistant runtime design.
• Ollama and OpenAI-compatible providers are relevant integrations because ZeroClaw routes model providers through its runtime.
• MCP-style tool and agent protocols are adjacent to ZeroClaw's channel/tool/gateway architecture.

Sources

• https://github.com/zeroclaw-labs/zeroclaw#readme
• https://github.com/zeroclaw-labs/zeroclaw/blob/master/docs/book/src/security/overview.md
• https://github.com/zeroclaw-labs/zeroclaw/blob/master/docs/book/src/tools/overview.md
• https://github.com/zeroclaw-labs/zeroclaw/wiki
• source_facts.package-manager