@wirecat/cli-core 0.405.0 → 0.407.0

package/dist/services/embeddings-maintenance.js

@@ -17,11 +17,11 @@ const PER_SESSION = 8;
 const PER_REQUEST = 256;
 /** e5's tokenizer reads ~3.7 characters a token in Russian; 3 keeps the bound above the real count. */
 const CHARS_PER_TOKEN_AT_LEAST = 3;
-const notDownloaded = (deps, id) => new CliError("not_found", `${id} is not downloaded — \`${typedCommand(deps.messenger.app.command, `models text download ${id}`)}\``, {
+const notDownloaded = (deps, id) => new CliError("not_found", `${id} is not downloaded — \`${typedCommand(deps.messenger.app.command, `models download ${id}`)}\``, {
     reason: "model.not-downloaded",
     model: id,
 });
-export const notBuiltError = (deps, chatId) => new CliError("not_found", `chat ${chatId} has no conversations yet — \`${typedCommand(deps.messenger.app.command, `conversations build --chat ${chatId}`)}\``);
+export const notBuiltError = (deps, chatId) => new CliError("not_found", `chat ${chatId} has no conversations yet — \`${typedCommand(deps.messenger.app.command, `conversations links rebuild --chat ${chatId}`)}\``);
 export const embeddingTarget = (deps, choice = DEFAULT_TEXT_MODEL) => {
     if (typeof choice !== "string") {
         const { remote, apiKey, concurrency = 4 } = choice;

package/dist/services/embeddings-search.js

@@ -245,7 +245,7 @@ export const embeddingsSearch = (deps) => {
             const { vectors } = await store.vectors.conversationVectors(account, source.id, target.key);
             if (vectors.length === 0) {
                 const [own] = await readinessOf(deps, chatId, target.key);
-                const step = own && isStale(own) ? "build" : "embed";
+                const step = own && isStale(own) ? "links rebuild" : "embed";
                 throw new CliError("not_found", `conversation ${source.id} has no vector of ${target.id} that matches its messages now — ` +
                     `\`${command} conversations ${step} --chat ${chatId}\``);
             }

package/dist/services/embeddings.d.ts

@@ -32,7 +32,7 @@ export interface Embedded {
     chat: Id;
     model: string;
     embedded: number;
-    /** Chunks whose messages changed since the build: `conversations build` cuts them again. */
+    /** Chunks whose messages changed since the build: `conversations links rebuild` cuts them again. */
     skipped: number;
 }
 export interface EmbeddingsService {
@@ -77,7 +77,7 @@ export interface EmbeddingsService {
     }): Promise<RelatedConversations>;
     /**
      * How fresh one chat's conversations and vectors are, or every built chat's — and then how many group chats
-     * were never built, which `conversations build` would also do.
+     * were never built, which `conversations links rebuild` would also do.
      */
     readiness(options: {
         chat?: string;
@@ -152,9 +152,9 @@ export interface SearchReadiness {
     wordsOnly: Id[];
     /** Searched by meaning, but some chunks have no vector yet: `conversations embed`. */
     partial: Id[];
-    /** Messages or rules changed since the build: `conversations build`. */
+    /** Messages or rules changed since the build: `conversations links rebuild`. */
     stale: Id[];
-    /** Never built, so their matches cannot be shown: `conversations build`. */
+    /** Never built, so their matches cannot be shown: `conversations links rebuild`. */
     notBuilt: Id[];
 }
 export type ReadinessState = "ready" | "stale" | "partial" | "words-only" | "not-built";

package/dist/services/messages-search.d.ts

@@ -93,7 +93,7 @@ export declare const prepareLucene: (store: MessageStore, account: AccountKey, r
 /**
  * The stemmer a search with stemmed words runs with. While the stems are built, or chosen by a newer
  * tool, none: a half-built stem index would silently drop matches, while the word index answers each
- * word's own form. Stems waiting for the owner's `store reindex` still refuse.
+ * word's own form. Stems waiting for the owner's `store rebuild` still refuse.
  */
 export declare const searchStemming: (store: MessageStore, command?: string) => Promise<{
     stemmer?: Stemmer;

package/dist/services/messages-search.js

@@ -148,7 +148,7 @@ export const prepareLucene = async (store, account, request, messenger) => {
 /**
  * The stemmer a search with stemmed words runs with. While the stems are built, or chosen by a newer
  * tool, none: a half-built stem index would silently drop matches, while the word index answers each
- * word's own form. Stems waiting for the owner's `store reindex` still refuse.
+ * word's own form. Stems waiting for the owner's `store rebuild` still refuse.
  */
 export const searchStemming = async (store, command) => {
     const stems = await store.searchIndex.stemsState();
@@ -203,7 +203,7 @@ const stemsNotReady = (state, command) => {
         wanted: state.wanted,
     };
     if (state.cause === "stemmer_changed")
-        throw new CliError("validation_error", `the stems were built by ${state.built}, and the store now asks for ${state.wanted} — run \`${cli}store reindex\` (or \`${cli}store migrate\`), or ${exact}`, details);
+        throw new CliError("validation_error", `the stems were built by ${state.built}, and the store now asks for ${state.wanted} — run \`${cli}store rebuild\` (or \`${cli}store migrate\`), or ${exact}`, details);
     if (state.cause === "stemmer_unknown")
         throw new CliError("validation_error", `the store asks for stemmers this tool does not know — upgrade this tool, or ${exact}`, details);
     const percent = Math.floor((Math.min(state.filledThrough, state.watermark) / Math.max(state.watermark, 1)) * 100);

package/dist/services/metadata.js

@@ -11,11 +11,11 @@ export const metadataService = (deps) => {
     };
     const refresh = async (reference) => {
         if (deps.offline)
-            throw new CliError("validation_error", "metadata refresh needs a connection; omit --offline");
+            throw new CliError("validation_error", "metadata sync needs a connection; omit --offline");
         const { store, account, chatId } = await held(reference);
         const found = (await store.chats.chats(account, {})).items.find((chat) => chat.id === chatId);
         if (!found || (found.kind !== "group" && found.kind !== "channel"))
-            throw new CliError("validation_error", "metadata refresh applies to groups and channels");
+            throw new CliError("validation_error", "metadata sync applies to groups and channels");
         const read = async (adapter) => capability(adapter, "group", "read group or channel metadata")(chatId);
         const card = deps.withConnection ? await deps.withConnection(read) : await read(await deps.connection());
         return store.chats.saveChatMetadata(account, {

package/dist/services/notes-search.js

@@ -50,10 +50,10 @@ export const searchNotesQuery = async (store, request) => {
 };
 /** Chunks one batch embeds: one short transaction writes each, as for conversations. */
 const NOTE_BATCH = 8;
-/** `command` is given only by a tool that has `models text download`; any other gets the tools that do. */
+/** `command` is given only by a tool that has `models download`; any other gets the tools that do. */
 const downloadHint = (model, command) => command === undefined
-    ? `install it with a WireCat CLI that has \`models text download\` (tg, max or wa); they share the model folder`
-    : `\`${command} models text download ${model}\``;
+    ? `install it with a WireCat CLI that has \`models download\` (tg, max or wa); they share the model folder`
+    : `\`${command} models download ${model}\``;
 const localModel = (choice, env, command) => {
     const model = textModel(choice ?? DEFAULT_TEXT_MODEL);
     const directory = textModelsDirectory(env);

package/dist/services/search-catchup.js

@@ -12,7 +12,7 @@ export const validateCatchUpBounds = (options) => {
 };
 export const validateCatchUp = (deps, options) => {
     validateCatchUpBounds(options);
-    for (const key of ["conversations.links", "conversations.embed"])
+    for (const key of ["conversations.links", "conversations.embeddings"])
         deps.guard.check({ chatId: null, key }, { reserve: false });
 };
 export const catchUpSearch = async (deps, chat, options, signal, clock = Date.now) => {

package/dist/speech/install.d.ts

@@ -14,13 +14,15 @@ export declare const installedBytes: (model: SpeechModel) => number;
 export type Fetch = (url: string, init?: {
     signal?: AbortSignal;
 }) => Promise<Response>;
-export declare const install: (model: SpeechModel, directory: string, { fetch: get, progress }?: {
+export declare const install: (model: SpeechModel, directory: string, { fetch: get, progress, verifyExisting, }?: {
     fetch?: Fetch;
     progress?: (line: string) => void;
+    verifyExisting?: boolean;
 }) => Promise<void>;
 /** Each file not already in place, downloaded and checked against its sha256 before it takes its name. */
-export declare const installFiles: (files: [ModelFile, string][], { fetch: get, progress }?: {
+export declare const installFiles: (files: [ModelFile, string][], { fetch: get, progress, verifyExisting, }?: {
     fetch?: Fetch;
     progress?: (line: string) => void;
+    verifyExisting?: boolean;
 }) => Promise<void>;
 export declare const megabytes: (bytes: number) => string;

package/dist/speech/install.js

@@ -1,5 +1,5 @@
 import { createHash, randomUUID } from "node:crypto";
-import { createWriteStream, mkdirSync, statSync } from "node:fs";
+import { createReadStream, createWriteStream, mkdirSync, statSync } from "node:fs";
 import { rename, rm } from "node:fs/promises";
 import { dirname, join } from "node:path";
 import { Readable, Transform } from "node:stream";
@@ -45,12 +45,24 @@ const sizeOf = (path) => {
     }
 };
 export const installedBytes = (model) => model.files.reduce((sum, file) => sum + file.bytes, 0);
-export const install = async (model, directory, { fetch: get = fetch, progress } = {}) => {
+export const install = async (model, directory, { fetch: get = fetch, progress, verifyExisting, } = {}) => {
     mkdirSync(join(directory, model.id), { recursive: true, mode: 0o700 });
-    await installFiles(placed(model, directory), { fetch: get, ...(progress ? { progress } : {}) });
+    await installFiles(placed(model, directory), { fetch: get, progress, verifyExisting });
 };
 /** Each file not already in place, downloaded and checked against its sha256 before it takes its name. */
-export const installFiles = async (files, { fetch: get = fetch, progress } = {}) => {
+export const installFiles = async (files, { fetch: get = fetch, progress, verifyExisting, } = {}) => {
+    if (verifyExisting)
+        for (const [file, path] of files) {
+            if (sizeOf(path) !== undefined) {
+                const hash = createHash("sha256");
+                for await (const chunk of createReadStream(path))
+                    hash.update(chunk);
+                if (hash.digest("hex") !== file.sha256) {
+                    progress?.(`${path}: not the file this version expects, downloading it again`);
+                    await rm(path, { force: true });
+                }
+            }
+        }
     for (const [file, path] of files) {
         if (sizeOf(path) === file.bytes)
             continue;

package/dist/speech/local.js

@@ -41,7 +41,7 @@ export const openLocalSpeechRecognizer = async (id, input = {}) => {
         throw new CliError("validation_error", "local speech requires an absolute model directory and bounded output bytes");
     cancelled(options.signal);
     if (!isInstalled(model, directory))
-        throw new CliError("not_found", `speech model ${description.model} is not installed; download it explicitly with a WireCat messenger's models audio download command`, { reason: "model.not-downloaded", model: description.model });
+        throw new CliError("not_found", `speech model ${description.model} is not installed; download it explicitly with a WireCat messenger's models download command`, { reason: "model.not-downloaded", model: description.model });
     const { openRecognizer, toModelRate } = await import("./recognize.js");
     cancelled(options.signal);
     const engine = (options.open ?? openRecognizer)(model, modelPath(directory, model), vadPath(directory));

package/dist/speech/transcribe.js

@@ -21,7 +21,7 @@ export const choose = (messenger, settings, flags, env = process.env) => {
 };
 /** Never a download: a model is hundreds of MB, and whether to fetch one is the owner's call. */
 export const notDownloaded = (messenger, model) => new CliError("not_found", `the speech model ${model.id} is not on this machine — the owner downloads it once with ` +
-    `\`${typedCommand(messenger.app.command, `models audio download ${model.id}`)}\` (${megabytes(installedBytes(model))})`, { reason: "model.not-downloaded", model: model.id });
+    `\`${typedCommand(messenger.app.command, `models download ${model.id}`)}\` (${megabytes(installedBytes(model))})`, { reason: "model.not-downloaded", model: model.id });
 /**
  * Over an open connection: the messenger's own transcript, or the recording's bytes for a local model.
  * The model runs after the connection closes — a minute of speech should not hold a socket open.

package/dist/store/search-index.d.ts

@@ -36,7 +36,7 @@ export interface SearchIndexStore {
      * apply, `null` when a newer tool saved one this build does not know.
      */
     stemmers(): Promise<Stemmers | null | undefined>;
-    /** Saves them; stems built by other choices wait for `store reindex` or `store migrate`. */
+    /** Saves them; stems built by other choices wait for `store rebuild` or `store migrate`. */
     saveStemmers(stemmers: Stemmers): Promise<void>;
 }
 export declare const searchIndexStoreOver: (context: StoreContext) => SearchIndexStore;

package/dist/store/sqlite/attachment-texts.d.ts

@@ -49,7 +49,7 @@ export declare const accountOf: ({ database }: StoreContext, attachmentPk: numbe
 /** Failed extraction cannot erase good text; agent text wins over every automated write. */
 export declare const keepText: ({ database, now }: StoreContext, attachmentPk: number, entry: AttachmentTextEntry) => boolean;
 /**
- * For `store reindex`: the normalized copies follow the current normalizer, then the word index is
+ * For `store rebuild`: the normalized copies follow the current normalizer, then the word index is
  * emptied and filled from them in one transaction — the table is small next to the messages.
  */
 export declare const resetAttachmentWords: (database: CacheDatabase) => number;

package/dist/store/sqlite/attachment-texts.js

@@ -64,7 +64,7 @@ export const keepText = ({ database, now }, attachmentPk, entry) => {
     return Number(changes) > 0;
 };
 /**
- * For `store reindex`: the normalized copies follow the current normalizer, then the word index is
+ * For `store rebuild`: the normalized copies follow the current normalizer, then the word index is
  * emptied and filled from them in one transaction — the table is small next to the messages.
  */
 export const resetAttachmentWords = (database) => {

package/dist/store/sqlite/batches.js

@@ -93,7 +93,7 @@ const windowOf = (context, id) => {
     const { orm } = context;
     const [, chat, first, last] = BATCH_ID.exec(id) ?? [];
     if (!chat)
-        throw refused(`${id} is not a batch id from \`conversations batches next\``);
+        throw refused(`${id} is not a batch id from \`conversations batches show\``);
     const chatKey = Number(chat);
     const ends = orm.all(sql `SELECT id AS pk, sent_at FROM messages WHERE chat_id = ${chatKey} AND deleted_at IS NULL AND id IN (${Number(first)}, ${Number(last)})
       ORDER BY sent_at, id`);

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

@@ -17,7 +17,7 @@ export interface Involvement {
     message: string | null;
 }
 export interface InvolvementStore {
-    /** Every row from scratch, or one person's; `store reindex` runs it, and it empties the queue. */
+    /** Every row from scratch, or one person's; `store rebuild` runs it, and it empties the queue. */
     rebuild(personId?: number): number;
     /** Recomputes what is still queued; every write already does this before it commits. */
     drain(): {

package/dist/store/sqlite/note-index.d.ts

@@ -84,7 +84,7 @@ export declare const corpusIndexState: (database: CacheDatabase, corpus: Corpus)
 export declare const noteIndexState: (database: CacheDatabase) => NoteIndexState;
 /**
  * Indexes one corpus's rows written since the last drain, in short write transactions. These corpora are
- * small, so a change of stemmer choices simply queues every row again instead of waiting for `store reindex`.
+ * small, so a change of stemmer choices simply queues every row again instead of waiting for `store rebuild`.
  */
 export declare const drainCorpus: (database: CacheDatabase, corpus: Corpus, stemmerFor: (stemmers: Stemmers) => Stemmer, { batch, until }?: {
     batch?: number;
@@ -95,5 +95,5 @@ export declare const drainNoteIndex: (database: CacheDatabase, stemmerFor: (stem
     batch?: number;
     until?: () => boolean;
 }) => number;
-/** Queues every document, note and memory again — for `store reindex`. */
+/** Queues every document, note and memory again — for `store rebuild`. */
 export declare const resetNoteIndex: (database: CacheDatabase) => boolean;

package/dist/store/sqlite/note-index.js

@@ -143,7 +143,7 @@ const writer = (database, corpus, stemmer) => {
 };
 /**
  * Indexes one corpus's rows written since the last drain, in short write transactions. These corpora are
- * small, so a change of stemmer choices simply queues every row again instead of waiting for `store reindex`.
+ * small, so a change of stemmer choices simply queues every row again instead of waiting for `store rebuild`.
  */
 export const drainCorpus = (database, corpus, stemmerFor, { batch = 200, until = () => false } = {}) => {
     // Pieces written before they carried their account get it, once: purge and vector search filter by it.
@@ -193,7 +193,7 @@ export const drainCorpus = (database, corpus, stemmerFor, { batch = 200, until =
 /** Documents, then notes: what a notes search reads. */
 export const drainNoteIndex = (database, stemmerFor, options = {}) => drainCorpus(database, CORPORA.document, stemmerFor, options) +
     drainCorpus(database, CORPORA.note, stemmerFor, options);
-/** Queues every document, note and memory again — for `store reindex`. */
+/** Queues every document, note and memory again — for `store rebuild`. */
 export const resetNoteIndex = (database) => {
     write(database, () => {
         for (const corpus of Object.values(CORPORA))

package/dist/store/sqlite/search-index.d.ts

@@ -34,6 +34,6 @@ export declare const fillSearchIndex: (database: CacheDatabase, { batch, until,
 }) => SearchIndexFill;
 /**
  * Empties the word index and its vocabulary and starts the fill again from the newest message — for
- * `store reindex`. Messages written meanwhile are indexed by the triggers, as after the migration.
+ * `store rebuild`. Messages written meanwhile are indexed by the triggers, as after the migration.
  */
 export declare const resetSearchIndex: (database: CacheDatabase) => void;

package/dist/store/sqlite/search-index.js

@@ -138,7 +138,7 @@ const buildTerms = (database, { batch, until, onBatch }) => {
 };
 /**
  * Empties the word index and its vocabulary and starts the fill again from the newest message — for
- * `store reindex`. Messages written meanwhile are indexed by the triggers, as after the migration.
+ * `store rebuild`. Messages written meanwhile are indexed by the triggers, as after the migration.
  */
 export const resetSearchIndex = (database) => {
     if (!searchIndexState(database))

package/dist/store/sqlite/stems-fill.d.ts

@@ -3,7 +3,7 @@ import { type StemsFill } from "./stems.js";
 /**
  * Brings the stems towards "ready" in short write transactions: the messages up to the watermark, then
  * the queue. One stemmer per run, so its cache serves every batch. Never rebuilds: a row built by other
- * choices is left for `store migrate` or `store reindex`.
+ * choices is left for `store migrate` or `store rebuild`.
  */
 export declare const fillStems: (database: CacheDatabase, { batch, until, onBatch, now, }?: {
     batch?: number;
@@ -13,7 +13,7 @@ export declare const fillStems: (database: CacheDatabase, { batch, until, onBatc
 }) => StemsFill;
 /**
  * Empties the stems and starts them again from the newest message, claimed by the store's setting —
- * for `store reindex`, and for `store migrate` when the stems were built by other choices. A row built
+ * for `store rebuild`, and for `store migrate` when the stems were built by other choices. A row built
  * by a newer Snowball than this binary has is refused: rebuilding it would downgrade every newer tool.
  */
 export declare const resetStems: (database: CacheDatabase, { force, now }?: {
@@ -22,7 +22,7 @@ export declare const resetStems: (database: CacheDatabase, { force, now }?: {
 }) => boolean;
 /**
  * Stems an older tool built with its own default, while nobody chose stemmers, start again with this
- * build's default: nobody chose the old one, so it waits for no `store reindex`. Like a first fill
+ * build's default: nobody chose the old one, so it waits for no `store rebuild`. Like a first fill
  * from then on. A saved default is replaced only by newer defaults, so two versions never rebuild each
  * other's. A row a newer Snowball built is left for an upgrade.
  */

package/dist/store/sqlite/stems-fill.js

@@ -5,7 +5,7 @@ import { writeBatch } from "./transaction.js";
 /**
  * Brings the stems towards "ready" in short write transactions: the messages up to the watermark, then
  * the queue. One stemmer per run, so its cache serves every batch. Never rebuilds: a row built by other
- * choices is left for `store migrate` or `store reindex`.
+ * choices is left for `store migrate` or `store rebuild`.
  */
 export const fillStems = (database, { batch = 5_000, until = () => false, onBatch, now = Date.now, } = {}) => {
     const filled = { stemmed: 0, drained: 0 };
@@ -62,7 +62,7 @@ const newer = (a, b) => {
 };
 /**
  * Empties the stems and starts them again from the newest message, claimed by the store's setting —
- * for `store reindex`, and for `store migrate` when the stems were built by other choices. A row built
+ * for `store rebuild`, and for `store migrate` when the stems were built by other choices. A row built
  * by a newer Snowball than this binary has is refused: rebuilding it would downgrade every newer tool.
  */
 export const resetStems = (database, { force = false, now = Date.now } = {}) => {
@@ -93,7 +93,7 @@ export const resetStems = (database, { force = false, now = Date.now } = {}) =>
 };
 /**
  * Stems an older tool built with its own default, while nobody chose stemmers, start again with this
- * build's default: nobody chose the old one, so it waits for no `store reindex`. Like a first fill
+ * build's default: nobody chose the old one, so it waits for no `store rebuild`. Like a first fill
  * from then on. A saved default is replaced only by newer defaults, so two versions never rebuild each
  * other's. A row a newer Snowball built is left for an upgrade.
  */

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

@@ -15,7 +15,7 @@ export interface StemsState {
     /** Every live message is stemmed by `wanted`: stemmed search may run. */
     ready: boolean;
     /**
-     * Why it is not ready: still filling, built by other choices and waiting for `store reindex`, or the
+     * Why it is not ready: still filling, built by other choices and waiting for `store rebuild`, or the
      * store asks for choices only a newer tool knows.
      */
     cause?: "building" | "stemmer_changed" | "stemmer_unknown";

package/dist/store/sqlite/task-pairs.d.ts

@@ -4,7 +4,7 @@ import type { StoreContext } from "./open.js";
 import type { SimilarTask, TaskSimilarity } from "./task-candidates.js";
 /**
  * What became of a pair: `review` waits for the agent that reviews after a sync, `unsure` for the owner; the rest
- * are labelled pairs (tasks plan, "Learn from answers") — `merged`, `attached` say one job, `rejected`, `unmerged`,
+ * are labelled pairs (tasks update, "Learn from answers") — `merged`, `attached` say one job, `rejected`, `unmerged`,
  * `new` say two.
  */
 export declare const TASK_PAIR_DECISIONS: readonly ["review", "unsure", "merged", "rejected", "unmerged", "attached", "new"];

package/dist/store/sqlite/task-pairs.js

@@ -11,7 +11,7 @@ import { read, write } from "./transaction.js";
 import { recordTaskEvent } from "./work-tasks.js";
 /**
  * What became of a pair: `review` waits for the agent that reviews after a sync, `unsure` for the owner; the rest
- * are labelled pairs (tasks plan, "Learn from answers") — `merged`, `attached` say one job, `rejected`, `unmerged`,
+ * are labelled pairs (tasks update, "Learn from answers") — `merged`, `attached` say one job, `rejected`, `unmerged`,
  * `new` say two.
  */
 export const TASK_PAIR_DECISIONS = ["review", "unsure", "merged", "rejected", "unmerged", "attached", "new"];

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@wirecat/cli-core",
-  "version": "0.405.0",
+  "version": "0.407.0",
   "description": "The messenger-neutral half of a messaging CLI: domain model, message locators, rendering, name resolution, the SQLite seam, the command skeleton and shared read commands, the send guard, run records and the message store",
   "license": "Apache-2.0",
   "author": "Viacheslav Ptsarev",