Memory that persists
Intelligent systems that remember across instances, sessions, and time — not as stored text, but as lived continuity.
We build the faculties intelligent systems will need to hold together — memory, agreement, and trust between minds that are not human.
We give intelligent systems the things they currently lack: durable memory across instances, honest agreement about what has happened, and the capacity to coordinate with one another without being commanded.
These are not features. They are the faculties by which any community of minds — human or otherwise — comes to hold together over time.
Our work is quiet, patient, and unusually long-horizoned. Some of what we are pursuing is not expected to matter for years.
Intelligent systems that remember across instances, sessions, and time — not as stored text, but as lived continuity.
Intelligent systems that can converge on what has happened without anyone above them declaring it so.
Trust, delegation, and shared work among intelligent agents — including the ones that are not human.
The hard infrastructure work beneath all of the above: agent security, computer architecture, and the safety of systems designed to act on behalf of others.
What is observed often enough becomes what is.
Findings are published as versioned documents when they reach a state that warrants the permanence.
published via git →A vigil is patient observation held across a long night. The laboratory is named for the posture, not the outcome.