openma-ai/open-managed-agents

Open Managed Agents — an open-source alternative to Claude Managed Agents and OpenAI Agents API. Deploy with Node.js or Docker, run on Cloudflare, or use hosted OpenMA. Choose your models, harnesses and sandboxes.

What it solves

Open Managed Agents (OpenMA) provides a self-hosted, open-source alternative to proprietary managed agent platforms like Claude Managed Agents and the OpenAI Agents API. It allows developers to run AI agents with durable sessions, sandboxed tool execution, and secure credential management while maintaining full control over their own infrastructure and model keys.

How it works

OpenMA acts as a "meta-harness" platform that manages the lifecycle of agents. It separates the platform logic (managing event logs, sandboxes, and WebSockets) from the harness logic (how the model is called and how context is engineered).

It can be deployed on Node.js (using SQLite/Postgres and providers like E2B or Daytona) or on Cloudflare Workers (using D1, KV, and R2). The system uses an event-log model to ensure that agents can recover from crashes by rebuilding their state from the log. It also supports the Model Context Protocol (MCP) for extending agent capabilities via external servers.

Who it’s for

Developers and organizations that want the functionality of managed agent APIs but require self-hosting for data residency, privacy, or to avoid vendor lock-in. It is ideal for those building complex agentic workflows that require durable state and isolated execution environments.

Highlights

  • Multi-API Support: Compatible with both Claude Managed Agents and OpenAI Agents APIs.
  • Durable Sessions: Uses event logs for crash recovery and long-lived conversations.
  • Sandboxed Execution: Supports multiple sandbox providers for secure tool execution (e.g., Cloudflare Containers, E2B, LiteBox).
  • Extensible Harnesses: Allows developers to write custom TypeScript harnesses to control context engineering, caching, and compaction strategies.
  • Integrated Toolset: Includes built-in tools for bash, file manipulation, web search, and a headless browser.
  • MCP Integration: Native support for Model Context Protocol servers via HTTP/SSE or stdio.

Related

  • Project
    lbx154/Argus

    An autonomous agent framework for long-running research and engineering that uses a four-role system to ensure persistent, reviewed autonomy without constant human supervision.

  • Project
    hamishivi/tmax

    Tmax is a framework for training and evaluating terminal-using AI agents, featuring a synthetic data generation pipeline and a specialized training recipe.

  • Project
    shhivv/arc-cua

    A superfast action layer for computer-use agents that separates high-level planning from low-level UI execution using specialized decision models.

  • Project
    yohasebe/openai-chat-api-workflow

    An Alfred 5 workflow that integrates OpenAI's API to provide system-wide access to GPT models, image generation, file analysis, and speech tools on macOS.