About MartinLoop

    We built MartinLoop because we wanted to stop babysitting our coding agents.

    AI finally made it possible to turn ideas into software without waiting on a large engineering organization. Then came the new problem: the more work we gave agents, the more time we spent supervising them.

    The breakthrough was being able to build.
    The frustration was still having to watch everything.

    MartinLoop keeps the job focused, bounded, checked, and accountable so you can give agents real responsibility and walk away.

    The Origin
    "The breakthrough was being able to build. The frustration was still having to watch everything."

    MartinLoop did not start as an abstract infrastructure idea. It came from trying to build real products with coding agents every day. One unattended run burned $65 overnight. Other runs drifted out of scope, retried the same failures, or created more review work than they saved.

    The lesson was simple: better models alone would not solve the problem. We needed a reliable system around the job — one that could keep the boundaries in force, check the work, explain failures, and tell us when the agent actually needed a person.

    MartinLoop is built using coding agents running through MartinLoop itself. The core is Apache 2.0 licensed because builders should be able to inspect and understand what sits around their agents.

    The Team

    Built by people who live this workflow every day.

    Long-time collaborators who use coding agents on real product work every day.

    Vakeesan Mahalingam, CFA

    Vakeesan Mahalingam, CFA

    Founder & CEO

    Repeat VC-backed founder and experienced operator. University of Waterloo Economics & Finance graduate and CFA charterholder. Former CI Global senior analyst/trader managing roughly $2B. Previously built Torram, raised $710K from Draper Associates, Boost VC, and Bitcoin Frontier Fund, secured a $300M tokenized bond MoU, and recruited 15 institutional validators. Today he uses coding agents heavily to turn ideas into products and built MartinLoop from that firsthand experience.

    Gobi Shanthan

    Gobi Shanthan

    Full-Stack Engineer

    Long-time technical collaborator and full-stack engineer with production experience across web2 and web3. Gobi focuses on hardening, testing, and turning MartinLoop into software that survives real-world use.

    Principles

    Trust should be earned.

    Trust should be earned.

    An agent saying “done” is not enough. The work should meet the checks you chose.

    Give agents responsibility, not unlimited freedom.

    Autonomy works better when the job has boundaries.

    Show what happened.

    When a run finishes or fails, you should not have to reconstruct the story yourself.

    What's next

    The code is live. The benchmark is public.
    Talk to us.

    Apache 2.0 licensed. If you're an investor, potential design partner, or just want to talk about trusting coding agents with real work — we want to hear from you.

    Backed by
    MEMBER OF
    NVIDIA Inception ProgramAWS ActivateGoogle for Startups

    MartinLoop is a member of the NVIDIA Inception Program, supporting startups building the future of AI.