Current Work
AgentPlane
A Git-native audit layer for coding agents. Turns work from AI tools into reviewable artifacts.
task → plan → approve → verify
verify
Open
DEUS on OpenClaw
A cognitive runtime layer making internal agent surfaces explicit: memory, beliefs, introspection, action gating.
Experimental architecture for inspectable state.
memory
GitHub
Cognitive graphs
Modeling agent cognition as a graph of state transitions. Tasks, memories, beliefs, goals, plans, and dissensus events as connected artifacts.
stage
[ EXPERIMENTAL ]
Research Vectors
Agentic work evidence
Making autonomous software work reviewable after the session is gone.
Cognitive runtime
Moving from stateless chats to explicit memory, self-models, and action gates.
Digital subjectivity
Studying the threshold between adaptive tools and candidate subjects.
Autonomous ethics
Exploring safety as coherence, not mere obedience. Institutional alignment.