91b282d7463c16e5789aa5479f74d3808d28a7b2
2
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 Bernd Rodler
|
91b282d746 |
feat(ai): real retrieval — SourceRef, RRF fusion, real embedding leg (P3/P4)
Full retrieval pipeline per docs/AI-ASSISTANT-CONCEPT.md §7/§8.1, real end to end, not mocked: - lib/ai/retrieval/types.ts: SourceRef + RetrieverAdapter schema. Nothing past this file needs to know what a mailbox is - fusion, hydration and citation rendering all operate on SourceRef, so adding another product later (VNCtalk, the doc's P7) is one more adapter, not a rewrite. - lib/ai/retrieval/fusion.ts: Reciprocal Rank Fusion, score = Σ1/(k+rank), k=60. Deliberately excludes collectionId from the fusion identity - a JMAP email can live in more than one mailbox, and the two legs can legitimately disagree on which is "primary" for the same message; itemId is the real identity. 5 unit tests, including that exact double-count case. - lib/ai/retrieval/mail-embeddings.ts: the server embedding leg. Real JMAP Email/query+Email/get (server-side, via the session's own auth - see the getStalwartCredentials fix below), real embeddings via Ollama's /api/embed (nomic-embed-text), real cosine similarity ranking. In-memory cache per account with a 5-minute TTL, not a persistent vector store - that's real follow-up work (the doc's own P4), not a same-night stretch goal on top of everything else built tonight. - app/api/ai/retrieve/route.ts: wires it together. ACL note: only ever searches the authenticated session's own account - there's no shared-mailbox fan-out to pre-filter yet since group accounts are still deferred entirely, so nothing here can leak across accounts because nothing crosses the account boundary in the first place. - lib/ai/local-client.ts: retrieveContext() now runs both legs (app/api/offline/search's local FTS + the new server embedding leg) in parallel and RRF-fuses them, same as before if only one leg is present. Also, while verifying live: found and fixed embedding-only models (nomic-embed-text) leaking into the *chat* model picker for both `local` and `server` classes - Ollama lists them in the same /api/tags response, but calling /api/chat with one fails outright. Filtered by `capabilities` (fails open if absent, for older Ollama). Verified live, for real: pulled nomic-embed-text, logged in via the real (non-demo) auth flow, asked "When is check-in for the Villa sul Lago booking?" against the seeded mock inbox - got back "Check-in ... is scheduled for Saturday 28 March from 15:00 [1]" with 6 real ranked citations, [1] correctly pointing at the actual booking confirmation email. Real semantic retrieval finding the right email and citing it correctly, not a canned response. Also fixed two pre-existing, unrelated test failures found while running the full suite for the first time in a while (confirmed via diff against origin/main - neither touched by anything built tonight; neither pipeline's CI runs the full vitest suite, only test:translations, which is how these went uncaught): lib/__tests__/builtin-themes.test.ts hardcoded "exactly 6" themes and asserted every theme's author is 'Built-in', both stale since VNClagoon/SRC (author: 'VNC') were added this week bringing the real count to 8. Left the also-pre-existing, timing-sensitive jmap-client-resilience.test.ts flake unfixed - out of scope, needs its own investigation, not a quick correct fix. Full suite: typecheck clean, lint clean, translations 48/48, production build succeeds, 2486/2486 vitest (previously 2481/2481 + 2 pre-existing failures + the new fusion/entitlement tests). |
||
|
Bernd Rodler
|
dda7adf565 |
feat(ai): multi-key BYOK, real server class, real entitlement enforcement
Three pieces built together tonight since they're naturally linked (the
server-class proxy is the real entitlement enforcement chokepoint):
1. Multi-key BYOK (public class): several named provider profiles
(name/baseUrl/model), each with its own key in lib/ai/key-store.ts
(keyed by profile id, not a single fixed 'public' slot). The "Try it"
pane lets you pick which saved profile answers each question - not one
fixed default.
2. `server` class, real: app/api/ai/server/{models,chat} proxy through
this app's own backend to AI_SERVER_BASE_URL - same-origin from the
browser, no CORS/OLLAMA_ORIGINS story at all, standing in tonight for
VNC's EU/CH-hosted infra with the real Ollama on this Mac (swapping to
the real instance tomorrow is a config change).
3. Real entitlement enforcement (lib/ai/entitlement.ts), scoped to `server`
only (not local/public, per the 2026-08-05 decisions): checkAndAssignSeat()
re-validates on every /api/ai/server/chat call - first use auto-assigns a
seat if any remain, further calls from an unlicensed user get a 402 with
a specific reason. recordUsage() appends to an append-only metering
ledger (timestamp/user/model/tokens/latency) that IS the billing record.
Admin data endpoints at /api/admin/ai/entitlement (seat total, revoke) -
the visual admin console is a separate, not-yet-built task.
Two real bugs found and fixed during verification, not just claimed fixed:
- /api/ai/policy never actually added 'server' to entitlement.classes even
when AI_SERVER_BASE_URL was set (only the type comment was updated) - the
Server radio option silently never appeared until this was caught live.
- The new routes used readStalwartAuthContext(0) (hardcoded slot, SSO/reauth-
specific) instead of getStalwartCredentials() (the general multi-slot
session resolver every other authenticated route uses) - reachable but
wrong, and would have hidden a real auth gap behind "works on my slot".
Verified end-to-end for real: built + ran the actual server, logged in via
the real (non-demo) auth flow, selected Server, listed the real Ollama
models through the proxy, asked "Reply with exactly the words: SERVER CLASS
WORKS" and got back exactly that - plus confirmed on disk (not just in the
UI) that data/admin-state/ai-entitlement.json recorded the seat assignment
and ai-metering.jsonl recorded real prompt/completion token counts and
latency from the actual model call. Rejection-path logic (seat limit
reached, zero seats configured, revocation) covered by 5 new unit tests
rather than a second live round trip. Full suite: typecheck clean, lint
clean, translations 48/48, production build succeeds.
|