Reference
Roadmap
An honest map of what works today versus what is designed but not yet shipped. Everything documented elsewhere reflects verified, live behavior.
Live today#
- REST API:
health,context,scopes,search,answer,units,units/:id,catalog/stats,ingest,audit,audit/verify,admin/seed. - Multimodal hybrid search: image/audio/video/doc across the text, visual, audio, and doc-visual planes; dense (vector index) + lexical (catalog FTS5), RRF fusion, cross-encoder rerank, sub-unit Locators, cost/latency trace.
- Grounded cited answers via
POST /v1/answer: retrieve → localize → verify → cite, with inline citations, Locators, and an NLI faithfulness score. - Security: RBAC/ABAC, scope clamping, grant intersection, per-hit ACL filtering, and a hash-chained audit log with independent verification (
GET /v1/audit/verify). - Developer surfaces: TypeScript + Python SDKs, the CLI, and the agent-native MCP server.
- Web console: sign-in, workspace overview, search playground, unit browser.
- Multimodal ingestion via
POST /v1/ingest(modality + plane + Locator).
Coming soon#
Not available yet
The following are part of the design but are not implemented today. Until they ship, use the REST API directly.
| Area | Status | What to use today |
|---|---|---|
| Grant / principal admin APIs | Planned | /v1/grants, /v1/principals return 404 today. |
| Namespace registry API | Planned | /v1/namespaces returns 404 today. |
| Raw source fetch | Planned | /v1/sources/:id returns 404 today. |
| SSO / Access (JWT) auth | Reserved | Use Bearer tokens or console sessions. |
API stability#
The contract is versioned under /v1. Response shapes documented here are stable for building against; treat similarity scores as informational (they shift as models change) and assert on invariants like ordering, membership, and ACL counts instead.