Control beyond
intelligence.

Our mission and vision: building towards safer AGI.

Porcelain plates frame a narrow cobalt passage

Our
mission.

Build the control layer that lets AI agents act with real authority, safely.

What an agent may do, what it actually did, and how it recovers when something goes wrong: checked, recorded and open to inspection, whichever model or agent is acting.

MELRA, our open-source control layer

Countless spheres pass through a cobalt X

Our
vision.

A future where increasingly capable intelligence stays under meaningful human control.

Safer AGI will not come from smarter models alone. It needs a limit and a record around every action, so people can hand real work to AI without handing over blind trust.

The questions we are working on

How we
get there.

The principles behind everything we build.

  1. Measured

    If we haven’t measured it, we don’t claim it. Failed experiments are part of the record too.

  2. Open

    You cannot trust a control layer you are not allowed to read. MELRA is open source.

  3. Lived

    We run what we build. Bhai, our personal agent, sends through MELRA, which checks for proof before calling a send done.

    Meet Bhai

Research, then infrastructure, then products, and back to research.