openma — self-hosted infrastructure for managed agents

What is openma?

openma (Open Managed Agents) is the infrastructure layer for AI agents. You bring the harness — the loop that decides what the model does next — and openma takes care of everything else: durable sessions, sandboxed code execution, tool calls, persistent memory, encrypted credentials, and crash recovery.

Run the Claude-compatible API on Node or Cloudflare, or use the official OpenAI SDK against a Node deployment. Both APIs use OpenMA’s durable session infrastructure, with distinct request and event contracts. Bring your own model keys through Model Cards. The hosted Console is also available for the Claude-compatible workflow.

See What’s new for the recent API, runtime, and memory updates.

Choose your path

I want to use openma.dev Sign up, create your first agent in the Console, connect Linear or GitHub. No deploy needed.

I want the Claude-compatible API REST endpoints, the CLI, custom skills, custom tools, MCP servers, and the harness interface.

I want to self-host Run Node with Docker and SQL storage, or deploy Workers and Containers to your Cloudflare account.

I want the OpenAI Agents API Connect the official OpenAI SDK to Node: sessions, function tools, subagents, streaming, and artifacts.

I want to contribute Repo layout, local dev loop, where the design RFCs live.

Why openma?

Most agent frameworks tell you how to build an agent — the loop, the prompt format, the orchestration shape. openma is different: it stays out of the harness and instead provides what every long-running agent needs but is painful to build.

Durable sessions

Persist session history in SQL on Node or Durable Objects on Cloudflare. Query committed events after reconnecting and continue conversations across restarts.

Sandboxed execution

Run tools through a local or remote sandbox provider. Choose host-side Pi execution or an ACP harness inside the sandbox; capabilities depend on the selected runtime.

Vault-backed credentials

Store credentials encrypted at rest and resolve MCP credentials through the control plane. Transparent sandbox egress enforcement depends on the provider; see the Runtime Host guide.

Built-in memory

Attach named Memory Stores across sessions. Agents use standard file tools under /mnt/memory, with version history and write-back on supported runtimes.

MCP, everywhere

Connect any MCP server and openma auto-generates mcp_* tools the model can call.

Three integrations, day one

Publish agents into Linear (assignable issues), GitHub (PRs), and Slack (channels) with OAuth-managed auth.

Next steps

Read the Concepts page A one-page tour of every noun: Agent, Session, Environment, Vault, Skill, Tool, Harness, Integration.

Run the Quickstart Sign up, get an API key, and create a session with curl in five minutes.