OpenLoop
Roadmap

An open platform for long-horizon AI agents that actually ship.

Supervised agents that plan, build, and hand back reviewable work. One engine, many roles: a researcher, a reviewer, a coder, and an assistant that coordinates them. Yours to download and run.

The path

NowReliability spine Building

Make the engine trustworthy enough to build a product on, the unglamorous work that everything else stands on.

  • Heartbeat liveness: a run is judged alive by its heartbeat, not a fixed clock, so a healthy long task is never false-killed.
  • Honest verification: the verifier drives the real app, tears it down cleanly, and works correctly across repositories.
  • Bounded self-healing: stuck work is caught and retried within limits, never looped forever.
Phase 2Download and run, in one click

Ship the dashboard as a native desktop app that embeds the engine: cross-platform installers and a first-run setup. No clone, no terminal, no config marathon. Install it, open it, give it a goal.

Phase 3Containerized coding

Every coding task runs inside its own sandbox container: isolated from the host, reproducible, and safe to run many at once. The execution model the serious agent platforms are built on, brought in cleanly.

Phase 4The unified spine

Migrate every role (researcher, reviewer, coder, assistant) onto one shared core with a self-healing supervisor, retiring the bolt-on parts. One runtime, many profiles.

Phase 5Become undeniable

Docs, templates, and a distribution push across the developer ecosystem. A serious agent platform that is genuinely easy to run earns its audience, and that audience is the whole point.

Down the line

Cloud and multi-device

Run it hosted, with agents working across many machines at once. Start a run from your phone, come back to a finished, reviewable change.

Teams

Shared workspaces, multiple operators, and roles, so a whole team can point the same agents at the same codebase.

Skill and agent marketplace

Share and install skills, tools, and agent profiles from the community. Capabilities grow without waiting on a core release.

Any model, anywhere

Route each role to the best model for the job, cloud or local, by cost, speed, and quality. No lock-in.

Memory that compounds

Every run teaches the system. It learns your codebase, your patterns, and your preferences, and gets sharper the more you use it.

Autonomous company mode

Point it at a goal, not a task. It plans, builds, reviews, and ships across many workstreams while you supervise the outcomes.

The horizon, not a commitment. These are the bets that make the near-term phases worth building.

Why now

The category is real and still wide open. The platforms that win are the ones developers can download, run, and trust on day one. The hard part is not a clever demo, it is reliability and a one-click start, and that is exactly what the roadmap above is ordered around.

The plan is simple: be the easiest serious agent platform to actually run, ship it where developers already are, and let distribution compound. That is the path the category leaders walked, and it is a walkable path.

1 repo
one system, downloadable
1 click
install and run, no setup
Self-hosted
runs on your machine, your keys

This roadmap is a living document. Order and scope shift as phases land.