ClawdHome English | 中文 Multi-Agent security console for macOS — run an entire AI Agent team safely on a single Mac. ClawdHome lets you run multiple independent AI Agent instances on one Mac — supporting both OpenClaw and Hermes Agent engines — with each instance isolated in its own macOS user account, runtime, data directory, and permission boundary. A single SwiftUI control panel backed by a privileged XPC helper…
Multi-Agent security console for macOS — run an entire AI Agent team safely on a single Mac.
ClawdHome lets you run multiple independent AI Agent instances on one Mac — supporting both OpenClaw and Hermes Agent engines — with each instance isolated in its own macOS user account, runtime, data directory, and permission boundary. A single SwiftUI control panel backed by a privileged XPC helper daemon covers initialization, monitoring, backup, model configuration, and IM channel onboarding, with no shell scripting required.
Most multi-Agent setups are either too weak (Chrome profiles share a single macOS user account — credentials and cookies are readable across profiles) or too heavy (VMs and Docker cannot run macOS-native desktop apps). ClawdHome fills the gap: strong isolation with low operational overhead.
Real isolation, enforced by the OS kernel: each Shrimp is a separate macOS user account. Processes, files, Keychain entries, and network policies are separated at the kernel level. If one Agent is compromised, its blast radius stops at its own UID boundary.
Browser-account isolation, not shared desktop Chrome: each Shrimp also gets its own managed browser-account layer, so Browser Bridge/OpenCLI profiles, CDP endpoints, cookies, and session metadata are shared only inside that Shrimp boundary, not with your main browser account or other Shrimps.
Safer privilege model: the UI never executes privileged operations directly. All system-level actions go through an explicit XPC helper (LaunchDaemon) with a typed, auditable call surface.
Dual-engine coexistence: run OpenClaw and Hermes Agent side-by-side on the same Mac, each independently configured, sharing a common API key store and backup system.
Unified operations: initialization wizard, gateway lifecycle, file management, diagnostics, config hot-reload, backup/restore, and Cron tasks all live in one panel — no custom scripts, no manual launchd wiring.
Local-first, compliance-friendly: all core features work offline. Data never leaves the machine unless you opt in to cloud services, which suits regulated industries with data residency requirements.
Highlights
Dual-engine multi-Agent: manage OpenClaw + Hermes Agents from a single panel. Each Shrimp supports multiple Agents, each with its own identity, model config, and IM binding.
Role Market and Skills Store: summon a preconfigured Agent team from the Role Market with one click; extend capabilities via the Skills Store without starting from scratch.
13+ IM channels out of the box: WeChat, Feishu, Telegram, Slack, WeCom, DingTalk, WhatsApp, email, and more — pair by QR scan or token form, unified channel directory maintained in one place.
All-in-one operations panel: health monitoring, watchdog auto-recovery, maintenance terminal, and an integrated diagnostics center covering environment, permissions, config, security, gateway, and network.
Layered backup and restore: per-Shrimp or global backups, restorable to any point in time.
Centralized model and provider management: API keys stored in an isolated Keychain, apply a model profile to multiple Agents in one action, supports custom providers and local model services.
English and Chinese localization: built on Stable.xcstrings, language follows the system setting.
Who It’s For
AI studios and independent builders: run Agents for multiple clients on one Mac, each client’s credentials and data physically separated. Watchdog keeps things running around the clock; backups give you a paper trail when clients ask for evidence.
Enterprise IT and compliance teams: fully on-premise deployment, no data leaves the machine by default. Gateway logs and config changes are traceable, fitting financial, medical, and legal scenarios with data sovereignty and audit requirements.
Developers and technical creators: clone an existing Shrimp as an experiment sandbox, test freely, delete when done — the main environment stays clean. Use a dedicated demo Shrimp for livestreams so your real API keys never appear on screen.
ClawdHome.app is the operator-facing control plane for status, setup, and day-to-day maintenance.
ClawdHomeHelper is the privileged boundary for user management, process control, file operations, installs, and system automation.
Each Shrimp runs as a separate macOS user with its own Agent runtime, data directory, and managed browser-account namespace.
Browser-account isolation
ClawdHome does not treat browser automation as “just another app on the host desktop”. Instead, each Shrimp gets a managed browser-account namespace used by both supported runtimes.
OpenClaw and Hermes can reuse the same login state for the same Shrimp.
Browser Bridge/OpenCLI selectors and CDP session metadata stay inside that Shrimp’s boundary.
Your real macOS browser profile is not the control surface exposed to Agents.
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?v=4)
评论