Governance layer for AI agents

Agents that know what they’re not allowed to do.

The problem: give an AI agent real tools and it will take real, irreversible actions — send the email, write the record, delete the row — sometimes on weak or unverified evidence.

The approach: check every action an agent proposes against an explicit policy before it runs, with scored provenance and an auditable, streamed trace of exactly what happened and why.

  • Policy gate evaluated before execution
  • Provenance scored 0–1 per source
  • Every decision streamed as an auditable trace

Governed Agents · Run Trace

Every action an agent proposes is checked against an explicit policy before it runs. This view is rendered entirely from the run’s trace event stream.

Next.js · server-side agent loop · streaming NDJSON trace · policy gate before executionIntervention console →Architecture & schema →Repo →

✓ allow
IdleLive calls a Claude model server-side and may cold-start on the first run; Sample is instant.

Run a task to watch the three agents coordinate and the governance gate allow or block the proposed action — streamed live from /api/run.