Pilot checklist & links
- Connect — sign in or demo session
- Ingest mission / synthetic corpus
- Query — optional raw recall
- Node viewer — inspect chunk / reflection JSON and provenance
- Chat — Guide then Avatar; Phone UI after
Better RAG — Military
Governed memory for this workspace
Ingest, Query, and Node viewer validate recall and structure; Chat exercises Guide and Avatar before the field Phone UI.
Workspace ID & technical notes
Workspace ID —
Snapshot —
Home
GET /api/health — sanity check (no auth).
Demo script (testers)
- Sign in (header). If prompted, pick a workspace — list from
GET /api/auth/workspaces. - Ingest first — same corpus for everyone in this workspace (
POST /api/v1/ingest). Then use Chat for Guide → Avatar verification. - Open Node viewer after ingest to confirm document roots, chunks, and reflection nodes (
GET /api/v1/nodes). - Open Chat and confirm
roleandindex.context_hits. Use Query for raw recall cards.
A large PDF can create many nodes at once; with role_manager.py thresholds (<10 Guide, 10–99 partial → Avatar today, 100+ complete) you may skip Guide immediately. For a predictable Guide-first pass, use a tiny fixture — we do not promise “always Guide first.”
Ingest
Load mission documents here, then continue in Chat to validate Guide → Avatar with your team.
API: POST /api/v1/ingest (field file). Reflection runs by default.
Query (semantic recall)
POST /api/v1/recall — scores and provenance from the API only (no fabricated hits).
Node viewer
Memory tree for this workspace — Document (ingest fingerprint) → Chunks (retrieval units) and Reflection (post-ingest synthesis when present). Inspect JSON loads the full node from the API.
Chat
Guide when memory is thin; Avatar for grounded answers. Phone UI is the downstream end-user shell.
API
POST /api/v1/avatar/chat — role, index.context_hits
Guide is active — add documents in Ingest, then return here.