Governed AI Research Agents

QLens 6 introduces a local-first agent coordinator for accelerating research without granting AI authority over canonical knowledge.

Governing boundary

Agents MAY collect candidates, group context, detect patterns, flag contradictions, draft reasoning candidates, and test repository policy. Agents MUST NOT create, edit, verify, supersede, or publish canonical research automatically.

Every run is stored under repository/agent-runs/ as a checksummed review bundle. Its canonical_write field is always false, and its state begins as awaiting_human_review.

Built-in agents

Agent	Responsibility	Canonical write
evidence	Rank existing repository objects related to a question	Never
context	Group objects by type within the question's Research Space	Never
pattern	Detect repeated tags and repository patterns	Never
contradiction	Surface contradiction edges and contested objects	Never
reasoning	Produce a candidate premise bundle, not a conclusion	Never
review	Detect unsupported reasoning and unsupported verified understandings	Never

Run lifecycle

Question → Agent run → Review bundle → Human examination → Governed proposal → Canonical approval

A future canonical change must still use the existing proposal/review transaction workflow.