Trinity is an executable control plane for AI-assisted work.
Human Owner
|
v
Trinity Kernel
|
+-- Planner (turn intent into scoped plan artifacts)
+-- Executor (run bounded work in approved scope)
+-- Verifier (check evidence and decide PASS/RETRY/NEEDS_HUMAN/DEAD)
+-- Audit Chain (append-only truth log)
+-- Capture Store (inputs, outputs, and evidence)
+-- Memory Index (retrieval over accepted artifacts)
Trinity does not trust claims. Trinity trusts artifacts.
An AI agent can propose, edit, and produce evidence. It cannot self-approve completion, bypass policy gates, or turn memory into truth.
- CLI kernel: deterministic commands in
.ai/cli/. - Session capsule: per-task working area under
.ai/sessions/<id>/. - State pointer:
.state/session_state.json, with canonicalgraph_stateand compatibility-onlylegacy_state. - Audit chain:
.ai/audit/events.ndjson, append-only and hash-linked. - Verifier: rule-driven checks that consume artifacts.
- Memory: retrieval index over accepted artifacts;
rrrusesmemory-cli index.
Detailed contracts live under docs/specs/. Operator docs are
practical entry points; the specs remain canonical for implementation
details.