ClawManager is a Kubernetes-native control plane for running AI agent instances for many users. It provisions and governs managed runtimes (OpenClaw, Hermes, OpenCode, and DeepSeek Harness), and users work through browser-based desktops while administrators keep visibility and policy control without exposing Kubernetes details. It targets platform teams hosting agent instances for multiple users, operators who need runtime visibility and command dispatch, and builders who want governed model access and reusable resource injection instead of per-instance setup.
Each runtime comes in two modes. Lite instances run through shared gateway runtime pools, and Pro instances keep dedicated desktop deployments for stronger isolation. An Agent Control Plane turns each instance into a managed runtime that registers, reports heartbeats and health, receives commands, and stays aligned with a desired state held by the platform. It deploys on k3s or Kubernetes with single-node or cluster storage profiles.
All model traffic passes through an AI Gateway compatible with OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages, which records audit traces, accounts for cost and token usage per instance and session, and applies risk rules that block or reroute requests. A separate Security Protection workspace covers runtime defense, host and container isolation, component trust, outbound governance, approvals, skill scanning, and an emergency circuit breaker.
Features
- Managed runtimes: OpenClaw, Hermes, OpenCode, and DeepSeek Harness workspaces in Lite (pooled) or Pro (dedicated) mode
- Instance lifecycle: resource presets, desktop and shell access, files, archives, password-protected share links, and batch operations
- AI Gateway: OpenAI- and Anthropic-compatible model entry point with audit, cost accounting, session usage, and risk rules
- Agent Control Plane: registration, heartbeat health, desired-state sync, and command dispatch for start, stop, config, and skill operations
- Team collaboration: a Leader agent plans, dispatches, verifies, and synthesises work across Worker agents, with templates, Team chat, and an execution kanban
- Resource management: reusable channels, skills, scheduled tasks, and resource packs injected at instance creation
- Skill Hub: a catalog of skills with ownership, versions, publishing, installation, and scanning
- Security Protection Platform: runtime defense, isolation, outbound governance, policy, approvals, full-chain audit, and a circuit breaker
- Northbound API: a separately deployed gateway with scoped tokens, owner isolation, asynchronous operations, rate limits, and mTLS
- Enterprise auth: LDAP and LDAPS sign-in with group-to-role mapping
