This project provides an open-source, self-hostable alternative to proprietary managed AI agent services, enabling developers to run autonomous AI agents via a standardized API. It targets developers building programmatic agent products, offering significant advantages in model and cloud flexibility, cost control, and data ownership compared to closed-source solutions.

How It Works

OpenClaw Managed Agents acts as a managed service layer built upon the OpenClaw agent framework. It deploys each active agent session within an isolated Docker container, orchestrated by a central service. This orchestrator exposes a four-primitive REST API (Agent, Environment, Session, Event) mirroring Claude Managed Agents, allowing applications to programmatically create, manage, and interact with AI agents. Key architectural choices include using SQLite for metadata persistence, SSE for event streaming, and WebSockets for real-time control, ensuring durable session state even across orchestrator restarts.

Quick Start & Requirements

• Primary install/run command: Clone the repository, set an API key environment variable (e.g., MOONSHOT_API_KEY), and run docker compose up --build -d.
• Non-default prerequisites: Docker, an API key for at least one supported LLM provider (e.g., Anthropic, OpenAI, Gemini, Groq).
• SDKs: Python, TypeScript, and OpenAI SDK drop-in compatibility are provided.
• Links: Examples are included within the README for API interaction and SDK usage.

Highlighted Details

• Model & Cloud Agnostic: Supports numerous LLM providers and can be deployed on any cloud or VPS with Docker, offering significant cost savings over managed services.
• Agent Versioning & Control: Agents are versioned immutably, with support for archiving and optimistic concurrency for updates. Per-session quotas (maxCostUsdPerSession, maxTokensPerSession, maxWallDurationMs) provide fine-grained cost management.
• Advanced Networking & Security: Features a limited networking mode with egress proxy and hostname allowlisting, enforced at the Docker bridge level for enhanced security.
• First-Class Subagents: Supports multi-agent delegation where child agents are first-class, inspectable sessions, with configurable depth caps and permission policies.
• OpenAI SDK Compatibility: Offers a drop-in replacement for the OpenAI API endpoint, including real per-token streaming.

Maintenance & Community

This project is described as new with no deployed customers yet. Specific details regarding maintainers, community channels (like Discord/Slack), or sponsorships are not provided in the README.

Licensing & Compatibility

• License Type: MIT.
• Compatibility: The MIT license permits commercial use and linking within closed-source applications.

Limitations & Caveats

As a new project, it lacks a production track record. The resource model is not yet a complete multi-tenant boundary, meaning agents, environments, and vaults are deployment-global rather than isolated per user. User-token authentication does not currently provide full tenant isolation.