Skip to main content

RAGs (Agent RAGs)

?view=agent-rags · Context: Corpus and skills

URL: http://localhost:8093/?view=agent-rags

RAG inventory

RAG detail

After upload

  1. Inventory of collections by host/profile (GET /agent-rags).
  2. Public library at top: upload + dry-run + ingest with INGEST confirmation.
  3. Each agent card: state (ok / missing / error), smoke search, Admin panel (upload / queue / drop) with rag:write.

Ingest step by step (UI)​

  1. Locate the profile row (e.g. ld-researcher on LDTeam Operaciones).
  2. Expand Admin (queue / drop).
  3. Upload document → choose .txt / .md / .pdf.
  4. Review queue (dry-run) → check chunks.
  5. Type INGEST → Ingest.

What to expect: after a successful ingest, state becomes ok and smoke search stops reporting collection unavailable.

If it fails: 504 on large PDFs (nginx timeout ~60 s). CLI alternative:

# dry-run
containers exec society-executor society rag ingest \
  --profile ld-researcher --host ldteam-operaciones \
  --path /path/mounted/in-the-executor

# real ingest
containers exec society-executor society rag ingest \
  --profile ld-researcher --host ldteam-operaciones \
  --path /path/mounted/in-the-executor --confirm INGEST

The executor image includes society-rag. If an older container reports ModuleNotFoundError: society_rag, rebuild and recreate only that service:

make build-executor
containers compose --env-file .env -f infra/containers-compose.yml \
  up -d --no-deps --force-recreate society-executor

Writes require SOCIETY_RAG_WRITE_ENABLED=true; the default is false. After ingest, society rag list --profile <p> --host <h> and this view must show the same collection.