Frequently asked questions

    Questions about trusting coding agents with real work.

    MartinLoop helps you delegate more software work to coding agents without supervising every step. Here are the practical questions people ask first. Also see the homepage, install guide, and the open-source repository.

    MartinLoop architecture diagram: the system between your agent and model providers
    How MartinLoop sits between your agent and any model.

    Can I actually leave my coding agent alone?

    That is the problem MartinLoop is designed to solve. You define the job and the checks; MartinLoop keeps the run inside configured limits and tells you whether it finished, stopped, or needs review.

    What is MartinLoop?

    MartinLoop is the system around coding-agent work. The agent writes the code. MartinLoop keeps the job bounded, checks the result, stops wasted runs, preserves recovery information, and gives you a clear record afterward.

    Is MartinLoop another coding agent?

    No. Keep Claude Code, Codex CLI, Gemini CLI, and other supported workflows. MartinLoop sits around the job rather than replacing the worker.

    Does MartinLoop make AI-generated code automatically safe?

    No. A VERIFIED outcome means the checks configured for that run passed. It does not guarantee the software is bug-free, secure, production-ready, or automatically safe to merge.

    How does MartinLoop keep agents accountable?

    The job has explicit boundaries and checks outside the agent's own completion claim. MartinLoop records the outcome, stop reason, failure class where applicable, and the evidence available for review.

    What happens when the agent gets stuck?

    Configured limits and stop conditions can end the run instead of letting it continue indefinitely. MartinLoop then tells you why it stopped and what needs attention.

    Can MartinLoop stop runaway spend?

    MartinLoop supports hard per-run budget limits. When the configured limit is reached, the run can stop instead of continuing to spend.

    What does VERIFIED mean?

    It means the configured verifier evidence passed for that run. It is a scoped claim about the checks you selected, not universal correctness.

    What are STOPPED and NEEDS REVIEW?

    STOPPED means a configured boundary or condition ended the run. NEEDS REVIEW means MartinLoop could not establish completion from the available evidence and a person should look.

    Is MartinLoop model-agnostic?

    Yes by architecture: the job, limits, checks, recovery, and outcome are separate from the worker underneath. Native coding-agent runtimes and OpenAI-compatible model endpoints are different execution paths, so only publicly released and validated paths are described as full coding-worker support.

    Is MartinLoop open source?

    Yes. MartinLoop Core is Apache 2.0 licensed. Hosted dashboard and team functionality are separate early-access services.

    Does MartinLoop replace code review or CI?

    No. MartinLoop works earlier and around the run. Your existing CI, review tools, Git workflow, and specialist security tools can stay.

    Explore MartinLoop