@wirecat/cli-core 0.551.0 → 0.552.0

package/dist/meetings/result-items.js

@@ -116,10 +116,17 @@ const embedding = v.union([
         }),
     }),
 ]);
+const repair = v.strictObject({
+    account_id: identifier,
+    applied: flag,
+    transcripts: count,
+    languages: v.array(v.strictObject({ language: nullableText, transcripts: count })),
+});
 export const MEETING_RESULT_ENTITIES = {
     meeting_ingestion: run,
     meeting_export: exported,
     meeting_association: meetingChange,
     meeting_transcription: transcription,
     meeting_embedding: embedding,
+    meeting_repair: repair,
 };

package/dist/meetings/store.d.ts

@@ -90,6 +90,11 @@ export interface SearchHit {
     startMs: number | null;
     /** Who spoke a transcript hit's cue, as the transcript names them; absent for chat and summary hits. */
     speakerName?: string | null;
+    /** A transcript hit's transcript language (`ru`, `en`…), `null` when unknown; absent for chat and summary hits. */
+    language?: string | null;
+    /** A transcript hit's line, from 1, of `linesTotal` lines in its transcript: where in the meeting it was said. */
+    line?: number;
+    linesTotal?: number;
     /** `corrected` when a query word matched only as a stored word recognition may have garbled. */
     match?: "words" | "corrected";
     /** For a `corrected` hit, the stored words each such query word matched in this passage. */
@@ -126,4 +131,21 @@ export interface MeetingStore {
     reviewMeetingLink?(meetingId: number, eventIds: number[], now: number): Promise<void>;
     cursor(accountId: number): Promise<string | null>;
     setCursor(accountId: number, value: string, now: number): Promise<void>;
+    /** Detects the language of the account's transcripts that have none; stores it only with `apply`. */
+    fillTranscriptLanguages?(input: {
+        accountId: number;
+        apply: boolean;
+        now: number;
+    }): Promise<TranscriptLanguageFill>;
+}
+export interface TranscriptLanguageFill {
+    accountId: number;
+    applied: boolean;
+    /** Transcripts that had no language. */
+    transcripts: number;
+    /** How many of them each language was found in; `null` where the text did not show one. */
+    languages: {
+        language: string | null;
+        transcripts: number;
+    }[];
 }

package/dist/meetings/testing/memory.js

@@ -121,6 +121,9 @@ export const memoryMeetingStore = () => {
                     text: r.text,
                     startMs: r.startMs,
                     speakerName: r.speakerName,
+                    language: t.transcript.language,
+                    line: r.position + 1,
+                    linesTotal: t.rows.length,
                 })));
                 rows.push(...d.chat.map((r) => ({
                     meetingId: d.meeting.id,

package/dist/meetings/transcription.d.ts

@@ -13,6 +13,8 @@ export interface MeetingMediaDescriptor {
         model: string;
         version?: string;
         timing?: "fixed-windows" | "recognition-segments";
+        /** The one language the model writes (`ru`); without it the language is read from the recognized text. */
+        language?: string;
     };
 }
 export interface MeetingRecognitionSegment {

package/dist/meetings/transcription.js

@@ -1,5 +1,6 @@
 import { CliError } from "@wirecat/cli-base";
 import { contentHash } from "./ingest.js";
+import { transcriptLanguage } from "./language.js";
 import { dispatchMeetingMutation } from "./meeting-management.js";
 const bounded = (value, max, name) => {
     if (!Number.isSafeInteger(value) || value < 1 || value > max)
@@ -46,6 +47,9 @@ export const transcribeMeeting = async (source, store, target, options = {}) =>
     const timing = recognition.timing ?? "recognition-segments";
     if (timing !== "fixed-windows" && timing !== "recognition-segments")
         throw new CliError("invalid_response", "unsupported recognition timing");
+    const modelLanguage = recognition.language;
+    if (modelLanguage !== undefined && (typeof modelLanguage !== "string" || !/^[a-z]{2,3}$/.test(modelLanguage)))
+        throw new CliError("invalid_response", "recognition language must be a two- or three-letter code");
     const media = Object.freeze({
         id: identifier(raw.id),
         contentHash: raw.contentHash,
@@ -137,7 +141,7 @@ export const transcribeMeeting = async (source, store, target, options = {}) =>
                             source: `local/${encodeURIComponent(media.recognition.engine)}/${encodeURIComponent(media.recognition.model)}`,
                             format: "local-transcription",
                         }),
-                    language: null,
+                    language: modelLanguage ?? transcriptLanguage(rows.map((row) => row.text)),
                     contentHash: hash,
                     externalCreatedAt: null,
                     metadata: provenance,

package/dist/search/lucene/types.d.ts

@@ -22,4 +22,11 @@ export declare const MESSAGE_PARSING: {
 /** Further apart than this, two words are rarely about one thing, even in a long message. */
 export declare const MOST_WORDS_APART = 50;
 export declare function searchError(reason: string, span: Span, alternative?: string): never;
+/**
+ * The search ran out of time, not the query out of bounds: a valid query over a large store. `timeout`, so a caller
+ * does not tell the person to fix what they typed.
+ */
+export declare function outOfTime(limitMs?: number): never;
+/** A search limit that leaves the answer without a part, such as its total, rather than failing it. */
+export declare const isQueryLimit: (error: unknown) => boolean;
 export declare function exhausted(budget: string): never;

package/dist/search/lucene/types.js

@@ -1,4 +1,4 @@
-import { CliError } from "@wirecat/cli-base";
+import { CliError, isCliError } from "@wirecat/cli-base";
 import { QUERY_LIMITS as PARSER_LIMITS, queryError } from "@wirecat/cli-base/query";
 export const QUERY_VERSION = 1;
 export const QUERY_LIMITS = {
@@ -22,6 +22,16 @@ export const MOST_WORDS_APART = 50;
 export function searchError(reason, span, alternative) {
     queryError(reason, span, alternative, SEARCH_QUERY);
 }
+/**
+ * The search ran out of time, not the query out of bounds: a valid query over a large store. `timeout`, so a caller
+ * does not tell the person to fix what they typed.
+ */
+export function outOfTime(limitMs = QUERY_LIMITS.milliseconds) {
+    throw new CliError("timeout", `search stopped at its time limit of ${limitMs / 1000}s — the query is valid; ` +
+        "add chat: or date: to search less of the store", { reason: "query.time-limit", budget: "time", limit_ms: limitMs, complete: false });
+}
+/** A search limit that leaves the answer without a part, such as its total, rather than failing it. */
+export const isQueryLimit = (error) => isCliError(error) && (error.details.reason === "query.too-large" || error.details.reason === "query.time-limit");
 export function exhausted(budget) {
     throw new CliError("validation_error", `search exceeded its ${budget} budget — narrow the chat or date range`, {
         reason: "query.too-large",

package/dist/services/messages-archive-search.js

@@ -1,5 +1,6 @@
-import { CliError, isCliError } from "@wirecat/cli-base";
+import { CliError } from "@wirecat/cli-base";
 import { parseLocator } from "../domain/locator.js";
+import { isQueryLimit } from "../search/lucene/types.js";
 import { TOTAL_CAP, totalOf } from "../search/total.js";
 import { discoveryRequest, searchDiscovery } from "./messages-discovery.js";
 import { countLucene, searchLucene, statsLucene } from "./messages-search.js";
@@ -53,7 +54,7 @@ export const searchStore = async (store, account, request, messenger = {}, meani
     const total = discovery
         ? totalOf(discovery.candidates, found.items.length, found.hasMore, !discovery.truncated)
         : totalOf(await countLucene(store, account, { ...(request.discover ? discoveryRequest(request) : request), limit: request.limit + offset }, messenger, TOTAL_CAP).catch((error) => {
-            if (isCliError(error) && error.details.reason === "query.too-large")
+            if (isQueryLimit(error))
                 return undefined;
             throw error;
         }), found.items.length, found.hasMore);

package/dist/services/messages-combined.js

@@ -4,7 +4,7 @@ import { parseLocator } from "../domain/locator.js";
 import { correctWords } from "../search/correct.js";
 import { PRESET_VERSION } from "../search/lucene/presets.js";
 import { FIELD_VERSION } from "../search/lucene/registry.js";
-import { exhausted, QUERY_LIMITS } from "../search/lucene/types.js";
+import { exhausted, outOfTime, QUERY_LIMITS } from "../search/lucene/types.js";
 import { normalize } from "../store/normalize.js";
 import { withExcerpt } from "./excerpt.js";
 import { coverageOf, prepareLucene } from "./messages-search.js";
@@ -31,7 +31,7 @@ export const searchCombined = async (store, account, request, messenger = {}, op
         if (request.signal?.aborted)
             throw new CliError("validation_error", "search was aborted", { reason: "query.aborted", complete: false });
         if (performance.now() - started > QUERY_LIMITS.milliseconds)
-            exhausted("milliseconds");
+            outOfTime();
     };
     check();
     const spans = request.ast === undefined && !request.exact && request.text !== undefined

package/dist/services/messages-discovery.js

@@ -3,7 +3,7 @@ import { implicitTextTerms, parseLucene, walkQuery } from "@wirecat/cli-base/que
 import { parseLocator } from "../domain/locator.js";
 import { fieldsOfQuery, meaningWords } from "../search/lucene/meaning-parts.js";
 import { validateFields } from "../search/lucene/registry.js";
-import { exhausted, MESSAGE_PARSING, QUERY_LIMITS } from "../search/lucene/types.js";
+import { exhausted, MESSAGE_PARSING, outOfTime, QUERY_LIMITS } from "../search/lucene/types.js";
 import { planQuestion, words } from "../search/question-plan.js";
 import { searchCombined } from "./messages-combined.js";
 import { prepareLucene, searchLucene } from "./messages-search.js";
@@ -101,7 +101,7 @@ export const searchDiscovery = async (store, account, asked, messenger = {}, mea
         if (request.signal?.aborted)
             throw new CliError("validation_error", "search was aborted", { reason: "query.aborted", complete: false });
         if (performance.now() - started > QUERY_LIMITS.milliseconds)
-            exhausted("milliseconds");
+            outOfTime();
     };
     const union = new Map();
     let truncated = false;

package/dist/services/notes-search.js

@@ -1,4 +1,4 @@
-import { CliError, isCliError } from "@wirecat/cli-base";
+import { CliError } from "@wirecat/cli-base";
 import { dateRange, parseLucene, timezoneOf } from "@wirecat/cli-base/query";
 import { INTERNAL_NOTES } from "../domain/providers.js";
 import { normalizeTag } from "../domain/tags.js";
@@ -9,7 +9,7 @@ import { entriesOfList, fitting } from "../search/account-list.js";
 import { fieldsOfQuery as fieldsOf, meaningWords } from "../search/lucene/meaning-parts.js";
 import { validateFields } from "../search/lucene/registry.js";
 import { hasStems } from "../search/lucene/resolved.js";
-import { SEARCH_QUERY } from "../search/lucene/types.js";
+import { isQueryLimit, SEARCH_QUERY } from "../search/lucene/types.js";
 import { createStemmer, DEFAULT_STEMMERS } from "../search/stem.js";
 import { TOTAL_CAP, totalOf } from "../search/total.js";
 import { openPictureEmbedderFor } from "./document-pictures.js";
@@ -286,12 +286,11 @@ const filterRefs = async (store, narrowing, scope) => {
         return found.hasMore ? undefined : found.refs;
     }
     catch (error) {
-        if (tooLarge(error))
+        if (isQueryLimit(error))
             return undefined;
         throw error;
     }
 };
-const tooLarge = (error) => isCliError(error) && error.details.reason === "query.too-large";
 /**
  * The notes the words find, with those found by meaning: every note the search would list, paged to the end. Never
  * below what the page already showed; a count the store refuses as too costly is that lower bound.
@@ -313,7 +312,7 @@ const notesTotal = async (store, text, { exact, scope, shown, hasMore, }, others
         });
     }
     catch (error) {
-        if (tooLarge(error))
+        if (isQueryLimit(error))
             return totalOf(undefined, shown, hasMore);
         throw error;
     }

package/dist/services/search-all-meetings.d.ts

@@ -22,6 +22,9 @@ export interface MeetingSearchItem {
     /** A transcript hit's cue, `meeting:<account>/<meeting>/<transcript>/<cue>`, which `meetings passage` opens. */
     cueRef?: string;
     speakerName?: string | null;
+    language?: string | null;
+    line?: number;
+    linesTotal?: number;
 }
 export interface MeetingSearchCursor {
     accountId: number;

package/dist/services/search-all-meetings.js

@@ -105,7 +105,12 @@ const collectMeetings = async (store, account, request, query) => {
                     id: hit.id,
                     startMs: hit.startMs,
                     ...(hit.scope === "transcript" && hit.ref !== undefined
-                        ? { cueRef: hit.ref, speakerName: hit.speakerName ?? null }
+                        ? {
+                            cueRef: hit.ref,
+                            speakerName: hit.speakerName ?? null,
+                            language: hit.language ?? null,
+                            ...(hit.line === undefined ? {} : { line: hit.line, linesTotal: hit.linesTotal }),
+                        }
                         : {}),
                 },
             });

package/dist/speech/local.d.ts

@@ -5,6 +5,8 @@ export interface LocalSpeechRecognitionModel {
     engine: "sherpa-onnx";
     model: string;
     version: string;
+    /** The one language the model writes; absent for a multilingual model. */
+    language?: string;
     sampleRate: 16000;
     maxChunkSamples: 480000;
 }

package/dist/speech/local.js

@@ -23,7 +23,14 @@ export const localSpeechRecognitionModel = (id) => {
         featureDim: model.featureDim,
     }))
         .digest("hex");
-    return Object.freeze({ engine: "sherpa-onnx", model: model.id, version, sampleRate: 16000, maxChunkSamples: 480000 });
+    return Object.freeze({
+        engine: "sherpa-onnx",
+        model: model.id,
+        version,
+        ...(model.language === undefined ? {} : { language: model.language }),
+        sampleRate: 16000,
+        maxChunkSamples: 480000,
+    });
 };
 const notInstalled = (id) => new CliError("not_found", `speech model ${id} is not installed; download it explicitly with a WireCat messenger's models download command`, { reason: "model.not-downloaded", model: id });
 /**

package/dist/speech/models.d.ts

@@ -18,6 +18,8 @@ export interface SpeechModel {
     id: string;
     title: string;
     languages: string;
+    /** The one language the model writes, so its transcripts need no guessing; absent for multilingual models. */
+    language?: string;
     files: ModelFile[];
     /** The part of sherpa-onnx's `modelConfig` that names the files, given where they are. */
     config: (path: (name: string) => string) => Record<string, unknown>;

package/dist/speech/models.js

@@ -8,6 +8,7 @@ export const MODELS = [
         id: "gigaam-v3",
         title: "GigaAM v3, with punctuation",
         languages: "Russian — the best of the three for Russian",
+        language: "ru",
         featureDim: 64,
         files: [
             {
@@ -45,6 +46,7 @@ export const MODELS = [
         id: "gigaam-v3-ctc",
         title: "GigaAM v3 CTC, with punctuation — a little faster",
         languages: "Russian — a little faster, rougher with capital letters",
+        language: "ru",
         featureDim: 64,
         files: [
             {

package/dist/store/driver.d.ts

@@ -23,7 +23,18 @@ export interface CacheDatabase {
     exec(sql: string): void;
     prepare(sql: string): CacheStatement;
     close(): void;
+    /**
+     * Set when the connection has `TIME_LIMIT_FUNCTION`. A statement runs to its end before any timer can fire, so a
+     * long one is stopped from inside: `WHERE … AND wc_time_limit(<epoch ms>)` throws `TimeLimitReached` once past it.
+     */
+    timeLimit?: true;
 }
+export declare const TIME_LIMIT_FUNCTION = "wc_time_limit";
+export declare class TimeLimitReached extends Error {
+    constructor();
+}
+/** The SQL function's body: 1 until `until` (epoch milliseconds) has passed. */
+export declare const beforeTimeLimit: (until: unknown) => number;
 export type OpenDatabase = (path: string) => CacheDatabase;
 /**
  * **Write-ahead mode, and a timeout rather than an instant failure.**

package/dist/store/driver.js

@@ -9,6 +9,19 @@
  * The surface is deliberately tiny. Everything above it is ordinary SQL, so replacing the driver —
  * or adding a third — is writing thirty lines, not learning an abstraction.
  */
+export const TIME_LIMIT_FUNCTION = "wc_time_limit";
+export class TimeLimitReached extends Error {
+    constructor() {
+        super("the statement ran past its time limit");
+        this.name = "TimeLimitReached";
+    }
+}
+/** The SQL function's body: 1 until `until` (epoch milliseconds) has passed. */
+export const beforeTimeLimit = (until) => {
+    if (Date.now() > Number(until))
+        throw new TimeLimitReached();
+    return 1;
+};
 /**
  * **Write-ahead mode, and a timeout rather than an instant failure.**
  *

package/dist/store/drivers/node-sqlite.d.ts

@@ -1,5 +1,5 @@
 import { DatabaseSync } from "node:sqlite";
-import type { CacheDatabase } from "../driver.js";
+import { type CacheDatabase } from "../driver.js";
 /**
  * Import this module only through a dynamic import that follows `quietSqliteWarning()`: a static import loads
  * `node:sqlite`, and on Node 22 its ExperimentalWarning, before the filter can run.

package/dist/store/drivers/node-sqlite.js

@@ -1,5 +1,6 @@
 // biome-ignore-all lint/plugin/no-sqlite-driver: cli-core owns the store, so it opens the driver itself
 import { DatabaseSync } from "node:sqlite";
+import { beforeTimeLimit, TIME_LIMIT_FUNCTION, } from "../driver.js";
 import { loadVectorExtension, vectorExtension } from "../vector-extension.js";
 /**
  * Import this module only through a dynamic import that follows `quietSqliteWarning()`: a static import loads
@@ -29,6 +30,7 @@ export const nodeClient = async (path, { readOnly = false } = {}) => {
 export const STATEMENT_CACHE_SIZE = 1000;
 export const cacheOverNodeSqlite = (database, cacheSize = STATEMENT_CACHE_SIZE) => {
     const statements = new Map();
+    database.function(TIME_LIMIT_FUNCTION, { deterministic: false, directOnly: true }, beforeTimeLimit);
     const prepare = (sql) => {
         const statement = database.prepare(sql);
         return {
@@ -38,6 +40,7 @@ export const cacheOverNodeSqlite = (database, cacheSize = STATEMENT_CACHE_SIZE)
         };
     };
     return {
+        timeLimit: true,
         exec: (sql) => database.exec(sql),
         prepare: (sql) => {
             const cached = statements.get(sql);

package/dist/store/sqlite/lucene.d.ts

@@ -1,6 +1,6 @@
 import type { Page } from "../../domain/models.js";
 import type { QueryExecution } from "../../search/lucene/resolved.js";
-import type { SqlValue } from "../driver.js";
+import { type SqlValue } from "../driver.js";
 import type { StoreContext } from "./open.js";
 import type { ScoredHit } from "./words.js";
 export { type Corpus, prefixOf } from "./lucene-fields.js";

package/dist/store/sqlite/lucene.js

@@ -1,7 +1,8 @@
 import { CliError } from "@wirecat/cli-base";
 import { compileAutomaton } from "../../search/lucene/automaton.js";
 import { isStemmed } from "../../search/lucene/resolved.js";
-import { exhausted, QUERY_LIMITS, searchError } from "../../search/lucene/types.js";
+import { exhausted, outOfTime, QUERY_LIMITS, searchError } from "../../search/lucene/types.js";
+import { TIME_LIMIT_FUNCTION, TimeLimitReached } from "../driver.js";
 import { emailHitsByPk } from "./emails.js";
 import { placeHits } from "./hit-places.js";
 import { combine, FIELD_COMPILERS, ftsMatch, SPEECH, } from "./lucene-fields.js";
@@ -54,7 +55,7 @@ export const countMatches = (context, execution, cap) => matchingPks(context, ex
  * The keys of the messages that match, no more than `cap + 1`: one SQL query, unranked. `undefined` when a leaf is
  * checked row by row (a preset, a pattern), where it would cost as much as the search itself.
  */
-export const matchingPks = (context, execution, cap) => {
+export const matchingPks = (context, execution, cap) => inTime(() => {
     const { database, check, where, from, withRanking, ranked, leaves } = compileQuery(context, {
         ...execution,
         newest: true,
@@ -66,6 +67,19 @@ export const matchingPks = (context, execution, cap) => {
         .all(...ranked.params, ...where.params, cap + 1);
     check();
     return rows.map(({ id }) => Number(id));
+});
+const timeLimited = (error) => {
+    if (error instanceof TimeLimitReached)
+        outOfTime();
+    throw error;
+};
+const inTime = (work) => {
+    try {
+        return work();
+    }
+    catch (error) {
+        return timeLimited(error);
+    }
 };
 const compileQuery = (context, execution, boundedAttachments = false) => {
     const { database } = context;
@@ -74,6 +88,7 @@ const compileQuery = (context, execution, boundedAttachments = false) => {
     if (mail && (execution.conversationIds || execution.conversationSince !== undefined))
         throw new CliError("validation_error", "mail has no conversations to search within");
     const started = context.now();
+    const until = Date.now() + QUERY_LIMITS.milliseconds;
     const budget = { work: 0 };
     let states = 0, expansions = 0;
     const patterns = new Map();
@@ -82,7 +97,7 @@ const compileQuery = (context, execution, boundedAttachments = false) => {
         if (execution.signal?.aborted)
             throw new CliError("validation_error", "search was aborted", { reason: "query.aborted", complete: false });
         if (context.now() - started > QUERY_LIMITS.milliseconds)
-            exhausted("time");
+            outOfTime();
     };
     const patternOf = (pattern, node) => {
         check();
@@ -197,7 +212,10 @@ const compileQuery = (context, execution, boundedAttachments = false) => {
                 exact: true,
             })), "OR"),
         ], "AND");
-    let where = combine([{ sql: "m.deleted_at IS NULL", params: [], exact: true }, scope, expression], "AND");
+    const timeLimit = database.timeLimit
+        ? [{ sql: `${TIME_LIMIT_FUNCTION}(?)`, params: [until], exact: true }]
+        : [];
+    let where = combine([{ sql: "m.deleted_at IS NULL", params: [], exact: true }, scope, expression, ...timeLimit], "AND");
     if (execution.conversationIds !== undefined || execution.conversationSince !== undefined) {
         const ids = execution.conversationIds;
         where = combine([
@@ -404,7 +422,7 @@ const compileQuery = (context, execution, boundedAttachments = false) => {
 /** The matcher and its consumer run synchronously in one read snapshot; exact keys stay in SQL. */
 export const withQuerySelection = (context, execution, consume) => {
     const { database } = context;
-    return read(database, () => {
+    return inTime(() => read(database, () => {
         const query = compileQuery(context, execution, true);
         const { check, leaves, where, from, withRanking, ranked } = query;
         check();
@@ -448,9 +466,17 @@ export const withQuerySelection = (context, execution, consume) => {
             throw new Error("query selection consumers must run synchronously");
         check();
         return result;
-    });
+    }));
+};
+const runQuery = (context, execution, by) => {
+    try {
+        return runCompiled(context, execution, by).catch(timeLimited);
+    }
+    catch (error) {
+        return timeLimited(error);
+    }
 };
-const runQuery = async (context, execution, by) => {
+const runCompiled = async (context, execution, by) => {
     const { corpus, rowsFrom, hits, database, check, grouped, where, from, withRanking, ranked, leaves, stemmed, exactTier, driver, baseFrom, filtered, ranking, relevance, rankedFrom, exactColumn, order, placeTier, projection, projectionParams, joinedFrom, evaluate, } = compileQuery(context, execution);
     if (by === "pks" && leaves.every(({ test }) => !test)) {
         const rows = database

package/dist/store/sqlite/meeting-hit-search.js

@@ -121,9 +121,9 @@ export function meetingHitSearchOver({ database }) {
                 if (after)
                     params.push(after.tier, after.tier, after.started, after.started, after.meeting, after.meeting, after.kind, after.kind, after.id);
                 const sql = `WITH matches AS (${matching}), matching AS (SELECT rid,tier FROM matches h WHERE NOT EXISTS(SELECT 1 FROM meeting_index_pending p WHERE p.id=h.rid/4 AND p.indexable_type=CASE h.rid%4 WHEN 1 THEN 'meeting_transcript_row' WHEN 2 THEN 'meeting_chat_message' ELSE 'meeting_summary' END)), selected AS (SELECT m.id,coalesce(m.started_at,0) AS started FROM meetings m WHERE ${clauses.join(" AND ")}), hits AS (
-        SELECT h.tier,m.started,m.id AS meeting_id,0 AS kind,r.id,r.start_ms,${bytes("r", ["text"])} AS bytes,t.id AS transcript_id,r.position FROM matching h JOIN meeting_transcript_rows r ON h.rid%4=1 AND r.id=h.rid/4 JOIN meeting_transcripts t ON t.id=r.meeting_transcript_id AND t.superseded_at IS NULL AND t.deleted_at IS NULL JOIN selected m ON m.id=t.meeting_id
-        UNION ALL SELECT h.tier,m.started,m.id,1,c.id,NULL,${bytes("c", ["text"])},NULL,NULL FROM matching h JOIN meeting_chat_messages c ON h.rid%4=2 AND c.id=h.rid/4 JOIN selected m ON m.id=c.meeting_id
-        UNION ALL SELECT h.tier,m.started,m.id,2,s.id,NULL,${bytes("s", fields)},NULL,NULL FROM matching h JOIN meeting_summaries s ON h.rid%4=3 AND s.id=h.rid/4 JOIN selected m ON m.id=s.meeting_id)
+        SELECT h.tier,m.started,m.id AS meeting_id,0 AS kind,r.id,r.start_ms,${bytes("r", ["text"])} AS bytes,t.id AS transcript_id,r.position,t.language,(SELECT count(*) FROM meeting_transcript_rows n WHERE n.meeting_transcript_id=t.id) AS lines_total FROM matching h JOIN meeting_transcript_rows r ON h.rid%4=1 AND r.id=h.rid/4 JOIN meeting_transcripts t ON t.id=r.meeting_transcript_id AND t.superseded_at IS NULL AND t.deleted_at IS NULL JOIN selected m ON m.id=t.meeting_id
+        UNION ALL SELECT h.tier,m.started,m.id,1,c.id,NULL,${bytes("c", ["text"])},NULL,NULL,NULL,NULL FROM matching h JOIN meeting_chat_messages c ON h.rid%4=2 AND c.id=h.rid/4 JOIN selected m ON m.id=c.meeting_id
+        UNION ALL SELECT h.tier,m.started,m.id,2,s.id,NULL,${bytes("s", fields)},NULL,NULL,NULL,NULL FROM matching h JOIN meeting_summaries s ON h.rid%4=3 AND s.id=h.rid/4 JOIN selected m ON m.id=s.meeting_id)
         SELECT h.*,${bytes("m", ["title"])}+${bytes("s", ["doc_url"])} AS context_bytes FROM hits h
         JOIN meetings m ON m.id=h.meeting_id LEFT JOIN meeting_summaries s ON h.kind=2 AND s.id=h.id
         ${seek} ORDER BY h.tier,h.started DESC,h.meeting_id,h.kind,h.id LIMIT ?`;
@@ -176,6 +176,13 @@ export function meetingHitSearchOver({ database }) {
                         id: Number(row.id),
                         text,
                         startMs: row.start_ms === null ? null : Number(row.start_ms),
+                        ...(kind === 0
+                            ? {
+                                language: row.language === null ? null : String(row.language),
+                                line: Number(row.position) + 1,
+                                linesTotal: Number(row.lines_total),
+                            }
+                            : {}),
                     };
                 });
                 const fresh = items.flatMap((item) => {

package/dist/store/sqlite/meeting-languages.d.ts

@@ -0,0 +1,4 @@
+import { type MeetingStore } from "../../meetings/index.js";
+import type { StoreContext } from "./open.js";
+export type MeetingLanguageStore = Required<Pick<MeetingStore, "fillTranscriptLanguages">>;
+export declare const meetingLanguagesOver: ({ database }: StoreContext) => MeetingLanguageStore;

package/dist/store/sqlite/meeting-languages.js

@@ -0,0 +1,36 @@
+import { transcriptLanguage } from "../../meetings/index.js";
+import { positiveReadInteger } from "./meeting-reads.js";
+import { read, writeBatch } from "./transaction.js";
+export const meetingLanguagesOver = ({ database }) => ({
+    async fillTranscriptLanguages({ accountId, apply, now }) {
+        positiveReadInteger(accountId, "accountId");
+        const found = read(database, () => {
+            const rows = database.prepare("SELECT text FROM meeting_transcript_rows WHERE meeting_transcript_id = ?");
+            return database
+                .prepare(`SELECT t.id FROM meeting_transcripts t JOIN meetings m ON m.id = t.meeting_id
+           WHERE m.account_id = ? AND t.language IS NULL AND t.deleted_at IS NULL ORDER BY t.id`)
+                .all(accountId)
+                .map(({ id }) => ({
+                id: Number(id),
+                language: transcriptLanguage(rows.all(Number(id)).map((row) => String(row.text))),
+            }));
+        });
+        if (apply) {
+            const update = database.prepare("UPDATE meeting_transcripts SET language = ?, updated_at = ? WHERE id = ? AND language IS NULL");
+            writeBatch(database, () => {
+                for (const { id, language } of found)
+                    if (language !== null)
+                        update.run(language, now, id);
+            });
+        }
+        const counts = new Map();
+        for (const { language } of found)
+            counts.set(language, (counts.get(language) ?? 0) + 1);
+        return {
+            accountId,
+            applied: apply,
+            transcripts: found.length,
+            languages: [...counts].map(([language, transcripts]) => ({ language, transcripts })),
+        };
+    },
+});

package/dist/store/sqlite/meetings.d.ts

@@ -1,5 +1,6 @@
 import type { Attachment, MeetingChatLinesStore, MeetingContextStore, MeetingParticipantSearchStore, MeetingPreparationContextStore, MeetingPreparationKnowledgeStore, MeetingPreparationNotesStore, MeetingRelatedHistoryStore, MeetingReviewStore, MeetingSearchIndexStore, MeetingSearchStore, MeetingSeriesHistoryStore, MeetingStatsStore, MeetingStore, MeetingTranscriptReceiptStore, MeetingTranscriptStore, NewRecord } from "../../meetings/index.js";
 import type { CacheDatabase } from "../driver.js";
+import { type MeetingLanguageStore } from "./meeting-languages.js";
 import { type MeetingContextReadCapabilities, type MeetingReadCapabilities } from "./meeting-reads.js";
 import type { StoreContext } from "./open.js";
 type Row = Record<string, unknown>;
@@ -11,7 +12,7 @@ export declare const saveAttachment: (database: CacheDatabase, type: string, par
 export declare const drainMeetingIndex: (database: CacheDatabase, batch?: number) => number;
 /** Each word of the query, as a prefix, all required. */
 export declare const wordsQuery: (query: string) => string | null;
-export type StoredMeetings = MeetingStore & MeetingPreparationContextStore & MeetingPreparationKnowledgeStore & MeetingPreparationNotesStore & MeetingRelatedHistoryStore & MeetingSeriesHistoryStore & MeetingTranscriptStore & MeetingReadCapabilities & MeetingContextReadCapabilities & MeetingParticipantSearchStore & MeetingSearchStore & MeetingTranscriptReceiptStore & MeetingContextStore & MeetingSearchIndexStore & MeetingStatsStore & MeetingReviewStore & MeetingChatLinesStore;
+export type StoredMeetings = MeetingStore & MeetingPreparationContextStore & MeetingPreparationKnowledgeStore & MeetingPreparationNotesStore & MeetingRelatedHistoryStore & MeetingSeriesHistoryStore & MeetingTranscriptStore & MeetingReadCapabilities & MeetingContextReadCapabilities & MeetingParticipantSearchStore & MeetingSearchStore & MeetingTranscriptReceiptStore & MeetingContextStore & MeetingSearchIndexStore & MeetingStatsStore & MeetingReviewStore & MeetingChatLinesStore & MeetingLanguageStore;
 /** The `MeetingStore` of `src/meetings/` over the shared store. */
 export declare const meetingStoreOver: (context: StoreContext) => StoredMeetings;
 export {};

package/dist/store/sqlite/meetings.js

@@ -11,6 +11,7 @@ import { meetingContextLinksOver } from "./meeting-context-links.js";
 import { meetingHitIndexOver } from "./meeting-hit-index.js";
 import { meetingHitSearchOver } from "./meeting-hit-search.js";
 import { INDEX_CODES, indexedText, meetingRowid, summaryOf } from "./meeting-index-values.js";
+import { meetingLanguagesOver } from "./meeting-languages.js";
 import { meetingParticipantSearchOver } from "./meeting-participant-search.js";
 import { meetingPreparationContextOver } from "./meeting-preparation-context.js";
 import { meetingPreparationKnowledgeOver } from "./meeting-preparation-knowledge.js";
@@ -516,17 +517,18 @@ const search = (database, query, filter) => {
             .prepare(`WITH hit AS (SELECT rowid AS r FROM meeting_words WHERE meeting_words MATCH ?),
        picked AS (SELECT id, row_number() OVER (ORDER BY coalesce(started_at, 0) DESC, id) AS n FROM (${picked.sql}))
        SELECT p.n, 0 AS kind, t.id AS a, r.position AS b, r.id, t.meeting_id, r.start_ms, coalesce(length(CAST(r.text AS BLOB)),0) AS bytes,
-              r.speaker_name AS speaker, m.account_id
+              r.speaker_name AS speaker, m.account_id, t.language,
+              (SELECT count(*) FROM meeting_transcript_rows n WHERE n.meeting_transcript_id = t.id) AS lines_total
          FROM hit h JOIN meeting_transcript_rows r ON h.r % 4 = 1 AND r.id = h.r / 4
          JOIN meeting_transcripts t ON t.id = r.meeting_transcript_id
            AND t.superseded_at IS NULL AND t.deleted_at IS NULL
          JOIN picked p ON p.id = t.meeting_id JOIN meetings m ON m.id = t.meeting_id
        UNION ALL
-       SELECT p.n, 1, c.id, 0, c.id, c.meeting_id, NULL, coalesce(length(CAST(c.text AS BLOB)),0), NULL, NULL
+       SELECT p.n, 1, c.id, 0, c.id, c.meeting_id, NULL, coalesce(length(CAST(c.text AS BLOB)),0), NULL, NULL, NULL, NULL
          FROM hit h JOIN meeting_chat_messages c ON h.r % 4 = 2 AND c.id = h.r / 4
          JOIN picked p ON p.id = c.meeting_id
        UNION ALL
-       SELECT p.n, 2, s.id, 0, s.id, s.meeting_id, NULL, coalesce(length(CAST(s.title AS BLOB)),0)+coalesce(length(CAST(s.overview AS BLOB)),0)+coalesce(length(CAST(s.sections AS BLOB)),0)+coalesce(length(CAST(s.next_steps AS BLOB)),0)+coalesce(length(CAST(s.content AS BLOB)),0), NULL, NULL
+       SELECT p.n, 2, s.id, 0, s.id, s.meeting_id, NULL, coalesce(length(CAST(s.title AS BLOB)),0)+coalesce(length(CAST(s.overview AS BLOB)),0)+coalesce(length(CAST(s.sections AS BLOB)),0)+coalesce(length(CAST(s.next_steps AS BLOB)),0)+coalesce(length(CAST(s.content AS BLOB)),0), NULL, NULL, NULL, NULL
          FROM hit h JOIN meeting_summaries s ON h.r % 4 = 3 AND s.id = h.r / 4
          JOIN picked p ON p.id = s.meeting_id
        ORDER BY 1, 2, 3, 4 LIMIT 10001`)
@@ -551,6 +553,9 @@ const search = (database, query, filter) => {
                         cuePosition: Number(row.b),
                     }),
                     speakerName: row.speaker === null ? null : String(row.speaker),
+                    language: row.language === null ? null : String(row.language),
+                    line: Number(row.b) + 1,
+                    linesTotal: Number(row.lines_total),
                 }
                 : {};
             return {
@@ -596,6 +601,7 @@ export const meetingStoreOver = (context) => {
         ...meetingParticipantSearchOver(context),
         ...meetingHitIndexOver(context),
         ...meetingContextLinksOver(context),
+        ...meetingLanguagesOver(context),
         async appendTranscriptsWithReceipt(input) {
             try {
                 return append(context, input);