Resources — Research

We publish what we learn. Open research, open protocols, open minds.

[ 01 ]Research areas

[A01]

Deterministic execution graphs

Making agent runs replayable artifacts — same graph, same inputs, same output, every time.

[A02]

Context-graph privacy

Query semantics that answer questions about a user's context without exporting the context itself.

[A03]

Agent cost attribution

Charging every token, tool call, and retry to the node that caused it — not the run that contained it.

[A04]

Trust-kernel audit trails

Append-only, user-legible records of every data access, with read budgets enforced at the kernel.

[ 02 ]Notes — current ledger

[R-001]Replayable runs: determinism contracts for agent execution graphs
[R-002]Answering without exporting: a query model for context-graph privacy
[R-003]Causal meters: per-node cost attribution for agent workloads
[R-004]Books a user can read: the trust kernel's audit trail
Collaborate with the lab