OpenMAO

OpenMAO

Build organizations, not just agents.

OpenMAO is an open-source substrate for AI-native organizations that run themselves, accountably. You define the mission, roles, goals, and guardrails. Agents do the work. The organization remembers what it learns, proposes corrections when patterns repeat, and earns more autonomy as it proves it can be trusted. The human role does not disappear; it rises from operator, to reviewer, to board.

Most autonomous-company demos hand a swarm of agents the keys and hope for the best. OpenMAO takes the opposite bet:

> Autonomy is earned, not assumed.

Every action an agent takes is owned, governed, and auditable, and the system widens what it is allowed to do only on a track record it can prove. Accountability is not the brake on autonomy; it is the road to it.

OpenMAO is open source and self-hostable. The organization you build, and everything it learns, belongs to you, not to a vendor and not to a cloud.

The Flywheel

OpenMAO is not a pile of features. It is one loop, and the loop is the product:

governance -> institutional memory -> self-correction -> self-learning
    ^                                                        |
    +--------- wider autonomy  hands out bounded work envelopes
    +-> agents, frameworks, and tools execute bounded tasks
    +-> risky actions route back through OpenMAO before side effects
    +-> outcomes return as organizational record

The boundary is the point: external frameworks may execute bounded tasks, but the work item, authority, approvals, memory consequences, event history, and world model live in OpenMAO. For anything that can send, spend, deploy, write, or mutate, OpenMAO sits in the execution path: the agent cannot reach the side effect without passing the gate. Swap your agent framework tomorrow and the organization, its memory, and its audit trail remain intact.

Reference adapters for frameworks such as LangGraph, CrewAI, OpenAI Agents SDK, or other runtimes are examples of this worker boundary, not OpenMAO's product category or source of truth.

The Autonomy Dial

Autonomy is not a switch. It is a dial that rises with earned trust (Organization.autonomy_level):

• advisory: humans do the work; OpenMAO suggests and records.
• supervised: agents act; humans approve each consequential action. This is where the local release begins.
• bounded: agents act within enforced limits; humans approve high-risk and out-of-bounds actions.
• board-governed: future horizon where agents run operations and humans govern at the policy level.

Every widening is earned on audited evidence, and every widening is reversible.

Contributing

OpenMAO is building toward something hard and worth it: organizations that can be trusted to run themselves. The governance substrate is the on-ramp; the self-correcting organization is the destination.

Start with [NORTHSTAR.md](NORTHSTAR.md). It is the first thing to read and it governs the project's direction. Then read [AGENTS.md](AGENTS.md) and [CONTRIBUTING.md](CONTRIBUTING.md) for how to work in the repo. The highest-leverage open work is the learning loop; the [roadmap](docs/ROADMAP.md) lays out the path from here to there.

Useful Commands

make lint
make format
make typecheck
make test
make api
make console
npm run cli -- approvals list
npm run cli -- approvals approve 
npm run cli -- approvals reject 
npm run cli -- events [run_id]
npm run cli -- learning scan
npm run cli -- learning proposals

Documentation

• [NORTHSTAR.md](NORTHSTAR.md) - why OpenMAO exists and where it is going. Start here.
• [docs/ARCHITECTURE.md](docs/ARCHITECTURE.md) - architecture and invariants.
• [docs/ROADMAP.md](docs/ROADMAP.md) - staged path from wedge to flywheel to autonomy dial.
• [docs/POSITIONING.md](docs/POSITIONING.md) - where OpenMAO fits in the wider ecosystem.
• [docs/LIMITATIONS.md](docs/LIMITATIONS.md) - known limitations, stated honestly, with the OWASP agentic-risk mapping.
• [docs/V0SCOPE.md](docs/V0SCOPE.md) - what the first release ships and what is deferred.
• [docs/VOCABULARY.md](docs/VOCABULARY.md) - canonical terms.
• [docs/DEPLOYMENTMODES.md](docs/DEPLOYMENTMODES.md) - local, managed, and enterprise shapes.
• [docs/examples/acmelearninglab.md](docs/examples/acmelearninglab.md) - default demo walkthrough.
• [CONTRIBUTING.md](CONTRIBUTING.md) - contributor workflow.
• [SECURITY.md](SECURITY.md) - security reporting and expectations.
• [GOVERNANCE.md](GOVERNANCE.md) - project governance.
• [CHANGELOG.md](CHANGELOG.md) - public release history.
• [LICENSE](LICENSE) - Apache-2.0 license.

Project Boundary

OpenMAO learns from public agent and orchestration projects and may integrate them through governed worker/capability adapters. It does not clone, vendor, fork, embed, or copy external framework code.

Framework adapters must stay interchangeable: if a team replaces one worker runtime with another, OpenMAO should still own the work, authority, approvals, memory promotion, audit trail, and world model.

Do not commit secrets, private data, local-only artifacts, or closed-source project material.

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: OpenMAO
• Source: OpenMAO/OpenMAO
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.