OpenClaw Managed Agents
github.com/stainlu/openclaw-managed-agents- Category
- AI Agents
- Rank
- No. 1096Tools index
- Pricing
- Open Source
- Type
- TOOL
- Use case
- Agent Building
- Interfaces
- API
- Builder
- @stainlu
- GitHub
- 440 stars
- Latest release
- v0.2.0
- Date
About
An open-source alternative to Claude Managed Agents that lets you run autonomous AI agents via REST API using any model from any provider. Deploy on your own infrastructure with full control over data and hosting.
What it does
It wraps OpenClaw in four managed concepts: agents, environments, sessions, and events. Each active session gets an isolated container, while SQLite metadata and JSONL history preserve conversations across container eviction. Clients send work over HTTP, stream results with server-sent events, and use WebSockets for controls such as cancellation and tool confirmation.
Why it's ranked here
The architecture tackles a real product-building gap with durable sessions, container isolation, versioned agent configuration, quotas, audit records, and observable child sessions. The main reservation is maturity: the project explicitly reports no deployed customers, while its operational model asks users to manage Docker capacity, persistence, credentials, and security settings.
What's good
Sessions survive compute replacement because conversation history lives outside ephemeral containers. The service supports immutable agent history with conflict detection, per-session cost, token, and duration limits, permission policies, structured audit records, Prometheus metrics, and restart recovery. Python, TypeScript, and OpenAI-compatible clients reduce integration friction.
Tradeoffs
A container per active session provides isolation but makes capacity planning central. Cold starts are expected when warming is disabled, while warm containers consume substantial idle CPU according to the source comments. Authentication is open when no token is configured. SQLite and host-mounted JSONL suit single-host deployment better than an immediately distributed control plane.
How to use it well
Choose it for a developer-built product that needs durable, inspectable agent sessions behind an application API. Start locally with Docker, keep warming off until latency demands justify its resource cost, then configure authentication, limits, persistence, and metrics before exposure. It does not replace the personal OpenClaw workflow centered on one operator and chat channels.
Technical notes+
src/index.ts assembles the Hono orchestrator, Docker runtime, SessionContainerPool, SQLite or memory store, JSONL reader, restart adoption, provider credential passthrough, and configurable active and warm pool limits. src/orchestrator/server.ts exposes REST, SSE, OpenAI-compatible chat completions, request IDs, metrics, audit hooks, and rate limiting. src/auth.ts supports an admin bearer token plus stored user tokens, but disables authentication when the deployment token is absent. src/audit.ts hashes bearer tokens for actor identity and treats audit writes as best effort. src/store/index.ts defaults to SQLite and can generate a persisted vault key for development. src/metrics.ts registers Node process, HTTP, pool, quota, boot, and session metrics. docs/architecture.md documents host-mounted JSONL event history, ephemeral session containers, optimistic agent versioning, warm pooling, and selective startup recovery. package.json requires Node.js 22.14 or newer and configures Vitest plus a Bash end-to-end test.
Observed
- License
- MIT
- Primary language
- TypeScript
- Runtime
- Node.js 22.14 or newer
- Deployment surface
- Docker Compose plus published orchestrator, agent runtime, egress proxy, and Telegram adapter images
- Client packages
- Python SDK on PyPI and TypeScript SDK on npm
- Interfaces
- HTTP REST, server-sent event streaming, WebSocket control, OpenAPI specification, and OpenAI-compatible chat completions
- Storage
- SQLite metadata by default, optional in-memory backend, and host-mounted JSONL session events
- Testing
- Vitest unit tests and a Bash end-to-end test command are configured
Read from README.md, package.json, src/index.ts, src/log.ts, src/auth.ts, src/audit.ts, src/metrics.ts, src/log.test.ts, src/auth.test.ts, src/rate-limit.ts, src/metrics.test.ts, src/rate-limit.test.ts, src/store/index.ts, src/orchestrator/server.ts, docs/architecture.md.
What it can do
Run autonomous AI agents via REST API
HTTP requests with agent instructions and parameters → Agent execution results and responses
Deploy AI agents on custom infrastructure
Docker containers and deployment configuration → Self-hosted agent service instances
Execute agents with any AI model from any provider
Model selection parameters and provider credentials → Agent responses using specified AI models
Isolate agent sessions using Docker containers
Agent execution requests → Isolated containerized agent environments
Provide Claude Managed Agents-compatible API interface
Claude-compatible API requests → Compatible API responses and data formats
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.