MARTINLOOP

MartinLoop — The OS for AI coding agents.

Claude and Codex can write the code. MartinLoop runs the system around them: rules before the run, checks before completion, a record after every run.

Open source · Apache 2.0 licensed · martin-loop on npm · Keesan12/Martin-Loop on GitHub

npm install -g martin-loop
martin run "your task" --budget 3 --verify "pnpm test"
martin inspect

AI coding is moving from autocomplete to autonomous work.

Agents read repos, edit files, run commands, retry tasks, and produce changes teams may merge. That changes the question. It is no longer just: can AI write code? It is: can we control, check, and trust the work it does?

The agent is not the system.

Claude writes. Codex works through the task. Cursor helps developers move faster. But teams still need one place to set the rules, check the result, stop runaway runs, and keep a record. That is MartinLoop.

Five things happen on every MartinLoop run.

  1. Smarter retries. Failed attempts compressed into structured signals — not dumped back raw. Token usage stays flat.
  2. Targeted failure handling. 11 failure classes, each with a different fix. Syntax errors get constraint repairs. Hallucinations get grounding checks.
  3. Hard budget cap. Set a dollar limit before the run starts. MartinLoop tracks spend in real time and stops before the cap.
  4. Smart exits. MartinLoop exits with a clean result when diminishing returns kick in or the budget ceiling approaches.
  5. Accurate cost accounting. Every token counted — including thinking tokens and sub-agent spend. No more under-reporting.

Pricing

  • Open Source Core — Free. Full governed runtime. Apache 2.0 licensed. CLI, local budget caps, JSONL run records, evidence-gated completions.
  • Developer Pro, Team, Growth, Business, Enterprise — Coming soon. Paid plans with hosted dashboard, team controls, and enterprise governance are in early access. Join the waitlist.

FAQ

What is MartinLoop?
The OS for AI coding agents. It gives teams rules, checks, and records around every Claude, Codex, or custom agent run.
Is this another coding agent?
No. Claude and Codex are the workers. MartinLoop is the system around the work they do.
Is MartinLoop open source?
Yes. The core is Apache 2.0 licensed. Hosted dashboard and managed control plane are commercial.
Who is it for?
Engineering teams, platform teams, and CTOs running AI coding agents in production who need control, auditability, and a record of what happened.