open-managed-agents
OpenMA handles sessions, execution and recovery, so you can build and run agents with any model, any harness, any sandbox, on any cloud. An open-source, self-ho
What is open-managed-agents?
OpenMA handles sessions, execution and recovery, so you can build and run agents with any model, any harness, any sandbox, on any cloud. An open-source, self-hostable alternative to Claude Managed Agents and OpenAI Agents API.
Overview of open-managed-agents
What it is
OpenMA (Open Managed Agents) is an open-source, self-hostable platform that handles sessions, execution, and recovery for building and running agents. It works with any model, harness, sandbox, and cloud infrastructure. It is positioned as an alternative to Claude Managed Agents and OpenAI Agents API.
Key features
- Durable sessions and recovery: preserves conversation history, streams progress, coordinates execution attempts without tying the session to one process, and supports workspace checkpoints and retained outputs.
- Reusable tools and integrations: includes shell, file, web, and browser tools; supports MCP services and Skills; and connects agents to platforms such as GitHub issues/PRs, Linear, Slack, and Feishu.
- Flexible stack: choose model (Anthropic, OpenAI, DeepSeek, OpenRouter, custom endpoints), harness (built-in Pi, Claude Code, Codex via ACP adapters), sandbox (E2B, Daytona, LiteBox, Cloudflare Sandbox), and backend services (database, queue, object storage) independently.
- Architecture: operates as a meta-harness, managing resources and lifecycle around an agent loop. Harness can run outside or inside the sandbox.
- Scaling: Node deployments can use shared PostgreSQL/MySQL and compatible storage across replicas; concurrent sessions are supported through durable execution outbox, renewable leases, and fencing.
- OpenMatter integration: connects work events to authorized context, agent turns, and recorded results; supports issue-to-PR workflows with application-defined permissions and approvals.
Use cases
Building and running agents that require durable sessions, cross-session recovery, and tool reuse. Deploying agents with custom model, harness, sandbox, and backend choices. Creating issue-to-PR workflows with feedback loops. Using managed execution for coding assistants, research assistants, or workplace automation integrating GitHub, Linear, Slack, or Feishu.
Getting started
See the repository README.
License and pricing
Licensed under Apache-2.0.
Based on the project's public README and repository metadata.