AI-Assisted Research¶

Run the full agent orchestra¶

qlens research-ai run QST-000001 --actor Mamun

The output is a path such as:

repository/agent-runs/arn-ab12cd34ef.json

Run selected agents¶

qlens research-ai run QST-000001 \
  --agent evidence \
  --agent contradiction \
  --agent review

Inspect runs¶

qlens research-ai list --json
qlens research-ai show ARN-AB12CD34EF --json
qlens research-ai verify

Review and accept useful work¶

An agent result is not canonical. Convert a useful recommendation into a governed patch proposal:

qlens propose \
  "Add reviewed evidence relationship" \
  '{"oid":"RSN-000001","set":{"evidence_oids":["EVD-000001"]}}'

qlens review list
qlens review approve PRP-XXXXXXXX \
  --reviewer Mamun \
  --note "Checked against the canonical evidence"

NUR¶

Run:

qlens build
mkdocs serve

Open Research Workspace → AI Research Review. NUR displays run identity, question, Research Space, participating agents, state, and creation time. It does not approve or commit results automatically.

Safety properties¶

No API key is required for deterministic repository agents.
No agent writes canonical objects.
Every run is checksummed.
Every run emits an immutable AIResearchRunCompleted event.
Acceptance still requires the proposal and human-review transaction path.