GameCult/Huginn
Watch
9
0
Fork
You've already forked Huginn
 0
Code Issues Pull requests 1 Projects Releases Packages Wiki Activity Actions

chat: search_chat over the shared source collection, huginn chat search #57

Merged
eureka merged 4 commits from gamecult-chat/chat-search into main 2026-11-11 19:11:25 +00:00
Conversation 0 Commits 4 Files changed 18 +1471 -81
eureka commented 2026-11-11 17:43:16 +00:00
Owner
Copy link

Cut gamecult-chat:cut_spec:cut-chat-search.r2: chat search in the source organ, read side only.

Operator: expect one forge re-embed at the deploy of this cut. SourcesMeta.corpus changes from forge:GameCult to forge:GameCult+chat:rooms and every point gains a corpus payload, so the existing meta-mismatch rule rebuilds huginn_sources once. Schedule the huginn-sources run; do not let it surprise a Soul pass. Until that rebuild runs, every forge and chat read (listing included) answers a clear stale-index error ("built for corpus forge:GameCult ... the rebuild is pending") instead of an empty page.

What it adds

  • SearchChat / Chat on huginn.mind_request.v1 / huginn.mind_response.v1 (schemas regenerated). ChatHit has no optional room id, event id or timestamp.
  • Selectors compose as AND: --room (id ! or alias #, any-of), --after/--before (span overlap; boundary hits trimmed and their span and ids recomputed), texts (0..8; none is a listing oldest first by ts_start_ms scroll, truncated when more exist).
  • One collection, one queue, one identity check and one fuse; a forge search now filters corpus=forge.
  • ChatPayload is the single owner of the chat point shape and upsert_chat its only write; nothing calls it until the ingest cut. Until then a chat search answers empty.
  • huginn chat search [TEXT]... [--room R]... [--after T] [--before T] [--top-k N] [--json].

Probe: deployed Qdrant is 1.17.1 (integer range indexes and scroll order_by exist).

Verification: Yggdrasil local container, cargo check --locked --workspace and cargo test --no-fail-fast --locked -p huginn-mind -p huginn-daemon -p eureka-state --features eureka-state/cli: only the 36 desk-door tests fail (they refuse uid 0, unrelated, same 36 as on main).

Disclosures

  • chrono moved from dev-dependency to dependency in eureka-state: the CLI parses --after/--before ages with it.
  • crates/huginn-daemon/src/index/tests.rs keeps the base file's mixed CRLF endings; its diff is content only.
  • Known limit (fix owned by cut chat-window; behaviour unchanged, documented in docs/chat-retrieval-map.md): hits are ordered by trimmed start, and a long chunk trimmed across rooms can reappear on the listing continuation.
  • Review fixes: ages parse on char boundaries and overflow is an error; CLI help no longer claims DM/Discord exclusion (the ingest side owns it); mutation rows for span bounds, scroll order/filter, nearest filter and the corpus/ts_end_ms indexes are caught (10/10).

Not here: the ingest job, the event-id window (cut chat-window).

🤖 Generated with Claude Code

Cut `gamecult-chat:cut_spec:cut-chat-search.r2`: chat search in the source organ, read side only. **Operator: expect one forge re-embed at the deploy of this cut.** `SourcesMeta.corpus` changes from `forge:GameCult` to `forge:GameCult+chat:rooms` and every point gains a `corpus` payload, so the existing meta-mismatch rule rebuilds `huginn_sources` once. Schedule the `huginn-sources` run; do not let it surprise a Soul pass. Until that rebuild runs, every forge and chat read (listing included) answers a clear stale-index error ("built for corpus forge:GameCult ... the rebuild is pending") instead of an empty page. What it adds - `SearchChat` / `Chat` on `huginn.mind_request.v1` / `huginn.mind_response.v1` (schemas regenerated). `ChatHit` has no optional room id, event id or timestamp. - Selectors compose as AND: `--room` (id `!` or alias `#`, any-of), `--after`/`--before` (span overlap; boundary hits trimmed and their span and ids recomputed), texts (0..8; none is a listing oldest first by `ts_start_ms` scroll, `truncated` when more exist). - One collection, one queue, one identity check and one `fuse`; a forge search now filters `corpus=forge`. - `ChatPayload` is the single owner of the chat point shape and `upsert_chat` its only write; nothing calls it until the ingest cut. Until then a chat search answers empty. - `huginn chat search [TEXT]... [--room R]... [--after T] [--before T] [--top-k N] [--json]`. Probe: deployed Qdrant is 1.17.1 (integer range indexes and scroll `order_by` exist). Verification: Yggdrasil local container, `cargo check --locked --workspace` and `cargo test --no-fail-fast --locked -p huginn-mind -p huginn-daemon -p eureka-state --features eureka-state/cli`: only the 36 desk-door tests fail (they refuse uid 0, unrelated, same 36 as on main). Disclosures - `chrono` moved from dev-dependency to dependency in `eureka-state`: the CLI parses `--after`/`--before` ages with it. - `crates/huginn-daemon/src/index/tests.rs` keeps the base file's mixed CRLF endings; its diff is content only. - Known limit (fix owned by cut chat-window; behaviour unchanged, documented in `docs/chat-retrieval-map.md`): hits are ordered by trimmed start, and a long chunk trimmed across rooms can reappear on the listing continuation. - Review fixes: ages parse on char boundaries and overflow is an error; CLI help no longer claims DM/Discord exclusion (the ingest side owns it); mutation rows for span bounds, scroll order/filter, nearest filter and the corpus/ts_end_ms indexes are caught (10/10). Not here: the ingest job, the event-id window (cut chat-window). 🤖 Generated with [Claude Code](https://claude.com/claude-code) 
Probe: the deployed Qdrant answers 1.17.1 at 127.0.0.1:6333 (>= 1.8: integer
range payload indexes and scroll order_by exist).

Wire: SearchChat / Chat, ChatQuery (door-checked by field), ChatHit with no
optional room_id, event ids or timestamps, published schemas regenerated.
Index: corpus payload on every point (forge searches must corpus=forge), chat
payload codec derived from the messages, keyword and integer indexes, upsert_chat
/ chat_nearest / chat_scroll on Qdrant and the fakes. CORPUS becomes
forge:GameCult+chat:rooms, so the existing meta mismatch rebuilds the forge index
once at deploy. sources::search_chat shares the identity check and fuse with
the forge search; chat rides the same queue (SourceAsk) and Found::Chat.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
`huginn chat search [TEXT]... [--room R]... [--after T] [--before T] [--top-k N] [--json]`;
T is an RFC 3339 instant or an age (6h, 2d), one parser. The render prints the
organ's header line per hit and each message verbatim with its event id, and a
truncated listing ends with the --after that continues it. chrono moves from
dev-dependency to dependency of eureka-state (already in the lock graph).

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
- A collection built for another corpus set is refused with a stale-index
  error on every read (listing included), not answered with an empty page.
- --after/--before ages parse on char boundaries and overflow is an error.
- Tests for inclusive span bounds, scroll order/filter, nearest filter and
  the corpus and ts_end_ms indexes; all ten surviving mutants now caught.
- index/tests.rs keeps the base file's CRLF endings (diff is content only).
- Drop the exclusion claim from the CLI help; document the listing limit.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
eureka merged commit 421ace5432 into main 2026-11-11 19:11:25 +00:00
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
GameCult/Huginn!57
No description provided.