@wirecat/cli-core 0.20.0 → 0.262.0

package/dist/domain/vocabulary.js

@@ -0,0 +1,34 @@
+import { CliError } from "@wirecat/cli-base";
+/** Search accent folding would make distinct names equal; canonical aliases retain their letters. */
+export const normalizeVocabularyName = (input) => {
+    if (typeof input !== "string" || input.length > 200 || /[\p{Cc}\p{Cf}]/u.test(input))
+        throw new CliError("validation_error", "a vocabulary name takes 1–200 characters without control characters");
+    const value = input.normalize("NFKC").toLowerCase().replace(/\s+/gu, " ").trim();
+    if (!value)
+        throw new CliError("validation_error", "a vocabulary name must not be empty");
+    return value;
+};
+/** Similarity is a review hint; only exact normalized names resolve canonical records. */
+export const vocabularySimilarity = (left, right) => {
+    const a = normalizeVocabularyName(left)
+        .replace(/[^\p{L}\p{N}]+/gu, " ")
+        .trim();
+    const b = normalizeVocabularyName(right)
+        .replace(/[^\p{L}\p{N}]+/gu, " ")
+        .trim();
+    if (!a || !b)
+        return 0;
+    if (a === b)
+        return 1;
+    const grams = (value) => {
+        const chars = Array.from(value);
+        return new Set(chars.slice(1).map((char, index) => `${chars[index]}${char}`));
+    };
+    const aa = grams(a);
+    const bb = grams(b);
+    if (aa.size + bb.size === 0)
+        return 0;
+    const intersection = [...aa].filter((gram) => bb.has(gram)).length;
+    return (2 * intersection) / (aa.size + bb.size);
+};
+//# sourceMappingURL=vocabulary.js.map

package/dist/domain/vocabulary.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"vocabulary.js","sourceRoot":"","sources":["../../src/domain/vocabulary.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,mBAAmB,CAAA;AAE5C,qGAAqG;AACrG,MAAM,CAAC,MAAM,uBAAuB,GAAG,CAAC,KAAa,EAAU,EAAE;IAC/D,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,MAAM,GAAG,GAAG,IAAI,iBAAiB,CAAC,IAAI,CAAC,KAAK,CAAC;QAClF,MAAM,IAAI,QAAQ,CAAC,kBAAkB,EAAE,qEAAqE,CAAC,CAAA;IAC/G,MAAM,KAAK,GAAG,KAAK,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC,WAAW,EAAE,CAAC,OAAO,CAAC,OAAO,EAAE,GAAG,CAAC,CAAC,IAAI,EAAE,CAAA;IAChF,IAAI,CAAC,KAAK;QAAE,MAAM,IAAI,QAAQ,CAAC,kBAAkB,EAAE,qCAAqC,CAAC,CAAA;IACzF,OAAO,KAAK,CAAA;AACd,CAAC,CAAA;AAED,0FAA0F;AAC1F,MAAM,CAAC,MAAM,oBAAoB,GAAG,CAAC,IAAY,EAAE,KAAa,EAAU,EAAE;IAC1E,MAAM,CAAC,GAAG,uBAAuB,CAAC,IAAI,CAAC;SACpC,OAAO,CAAC,kBAAkB,EAAE,GAAG,CAAC;SAChC,IAAI,EAAE,CAAA;IACT,MAAM,CAAC,GAAG,uBAAuB,CAAC,KAAK,CAAC;SACrC,OAAO,CAAC,kBAAkB,EAAE,GAAG,CAAC;SAChC,IAAI,EAAE,CAAA;IACT,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC;QAAE,OAAO,CAAC,CAAA;IACtB,IAAI,CAAC,KAAK,CAAC;QAAE,OAAO,CAAC,CAAA;IACrB,MAAM,KAAK,GAAG,CAAC,KAAa,EAAE,EAAE;QAC9B,MAAM,KAAK,GAAG,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,CAAA;QAC/B,OAAO,IAAI,GAAG,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,KAAK,CAAC,KAAK,CAAC,GAAG,IAAI,EAAE,CAAC,CAAC,CAAA;IAC/E,CAAC,CAAA;IACD,MAAM,EAAE,GAAG,KAAK,CAAC,CAAC,CAAC,CAAA;IACnB,MAAM,EAAE,GAAG,KAAK,CAAC,CAAC,CAAC,CAAA;IACnB,IAAI,EAAE,CAAC,IAAI,GAAG,EAAE,CAAC,IAAI,KAAK,CAAC;QAAE,OAAO,CAAC,CAAA;IACrC,MAAM,YAAY,GAAG,CAAC,GAAG,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,EAAE,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,MAAM,CAAA;IAClE,OAAO,CAAC,CAAC,GAAG,YAAY,CAAC,GAAG,CAAC,EAAE,CAAC,IAAI,GAAG,EAAE,CAAC,IAAI,CAAC,CAAA;AACjD,CAAC,CAAA"}

package/dist/embeddings/child.d.ts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=child.d.ts.map

package/dist/embeddings/child.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"child.d.ts","sourceRoot":"","sources":["../../src/embeddings/child.ts"],"names":[],"mappings":""}

package/dist/embeddings/child.js

@@ -0,0 +1,19 @@
+import { createInterface } from "node:readline";
+import { openEmbedder } from "./embed.js";
+const { model, directory, threads } = JSON.parse(process.argv[2] ?? "{}");
+const embedder = await openEmbedder(model, directory, { threads });
+// stdout carries the answers alone, one JSON line each: anything else there breaks the parent's reading.
+const answer = (line) => process.stdout.write(`${JSON.stringify(line)}\n`);
+answer({ ready: true });
+for await (const line of createInterface({ input: process.stdin })) {
+    const { id, texts, kind } = JSON.parse(line);
+    try {
+        answer({ id, vectors: (await embedder.embed(texts, kind)).map((vector) => Array.from(vector)) });
+    }
+    catch (error) {
+        answer({ id, error: error instanceof Error ? error.message : String(error) });
+    }
+}
+await embedder.close();
+process.exit(0);
+//# sourceMappingURL=child.js.map

package/dist/embeddings/child.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"child.js","sourceRoot":"","sources":["../../src/embeddings/child.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,eAAe,EAAE,MAAM,eAAe,CAAA;AAC/C,OAAO,EAAa,YAAY,EAAE,MAAM,YAAY,CAAA;AAGpD,MAAM,EAAE,KAAK,EAAE,SAAS,EAAE,OAAO,EAAE,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,IAAI,CAIvE,CAAA;AACD,MAAM,QAAQ,GAAG,MAAM,YAAY,CAAC,KAAK,EAAE,SAAS,EAAE,EAAE,OAAO,EAAE,CAAC,CAAA;AAClE,yGAAyG;AACzG,MAAM,MAAM,GAAG,CAAC,IAAY,EAAE,EAAE,CAAC,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;AAClF,MAAM,CAAC,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAA;AACvB,IAAI,KAAK,EAAE,MAAM,IAAI,IAAI,eAAe,CAAC,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,EAAE,CAAC,EAAE,CAAC;IACnE,MAAM,EAAE,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAgD,CAAA;IAC3F,IAAI,CAAC;QACH,MAAM,CAAC,EAAE,EAAE,EAAE,OAAO,EAAE,CAAC,MAAM,QAAQ,CAAC,KAAK,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,EAAE,CAAC,CAAA;IAClG,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,MAAM,CAAC,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC,CAAA;IAC/E,CAAC;AACH,CAAC;AACD,MAAM,QAAQ,CAAC,KAAK,EAAE,CAAA;AACtB,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAA"}

package/dist/embeddings/embed.d.ts

@@ -0,0 +1,36 @@
+import type { TextModel } from "./models.js";
+export declare const textModelsDirectory: (env?: NodeJS.ProcessEnv) => string;
+export declare const placedText: (model: TextModel, directory: string) => [import("../speech/models.js").ModelFile, string][];
+export declare const isTextModelInstalled: (model: TextModel, directory: string) => boolean;
+/** Past 8 threads one session gains nothing; Bun takes 1 unless told (`2026-10-02-embeddings.md`). */
+export declare const defaultThreads: () => number;
+export type Kind = "query" | "passage";
+export interface Embedder {
+    readonly model: TextModel;
+    /** One unit vector per text, in order. */
+    embed(texts: string[], kind: Kind): Promise<Float32Array[]>;
+    close(): Promise<void>;
+}
+/**
+ * Models a long-running process keeps open between searches — the MCP server, where loading one costs
+ * ~1 s a call. One no search has used for `idleMs` is closed, and the next search loads it again: e5-small
+ * holds ~1 GB, for as long as an agent's session lasts otherwise. Whoever makes it closes it; a one-shot
+ * command never makes one.
+ */
+export interface WarmEmbedders {
+    get(key: string, open: () => Promise<Embedder>): Promise<Embedder>;
+    close(): Promise<void>;
+}
+export declare const WARM_IDLE_MS: number;
+export declare const warmEmbedders: ({ idleMs }?: {
+    idleMs?: number;
+}) => WarmEmbedders;
+/** A text longer than the model's limit keeps its beginning and the closing special token. */
+export declare const truncated: (ids: number[], max: number) => number[];
+export declare const unit: (vector: Float32Array) => Float32Array;
+/** The mean over a single text's tokens: one text a run, so no padding to mask. */
+export declare const meanOf: (data: Float32Array, tokens: number, dims: number) => Float32Array;
+export declare const openEmbedder: (model: TextModel, directory: string, { threads }?: {
+    threads?: number;
+}) => Promise<Embedder>;
+//# sourceMappingURL=embed.d.ts.map

package/dist/embeddings/embed.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"embed.d.ts","sourceRoot":"","sources":["../../src/embeddings/embed.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,aAAa,CAAA;AAE5C,eAAO,MAAM,mBAAmB,SAAS,MAAM,CAAC,UAAU,KAAiB,MAA4C,CAAA;AAEvH,eAAO,MAAM,UAAU,UAAW,SAAS,aAAa,MAAM,wDACoC,CAAA;AAElG,eAAO,MAAM,oBAAoB,UAAW,SAAS,aAAa,MAAM,KAAG,OAC/B,CAAA;AAE5C,sGAAsG;AACtG,eAAO,MAAM,cAAc,QAAO,MAA6C,CAAA;AAE/E,MAAM,MAAM,IAAI,GAAG,OAAO,GAAG,SAAS,CAAA;AAEtC,MAAM,WAAW,QAAQ;IACvB,QAAQ,CAAC,KAAK,EAAE,SAAS,CAAA;IACzB,0CAA0C;IAC1C,KAAK,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,IAAI,EAAE,IAAI,GAAG,OAAO,CAAC,YAAY,EAAE,CAAC,CAAA;IAC3D,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC,CAAA;CACvB;AAED;;;;;GAKG;AACH,MAAM,WAAW,aAAa;IAC5B,GAAG,CAAC,GAAG,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,OAAO,CAAC,QAAQ,CAAC,GAAG,OAAO,CAAC,QAAQ,CAAC,CAAA;IAClE,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC,CAAA;CACvB;AAED,eAAO,MAAM,YAAY,QAAc,CAAA;AAEvC,eAAO,MAAM,aAAa,gBAA+B;IAAE,MAAM,CAAC,EAAE,MAAM,CAAA;CAAE,KAAQ,aAoDnF,CAAA;AAoBD,8FAA8F;AAC9F,eAAO,MAAM,SAAS,QAAS,MAAM,EAAE,OAAO,MAAM,KAAG,MAAM,EACe,CAAA;AAE5E,eAAO,MAAM,IAAI,WAAY,YAAY,KAAG,YAK3C,CAAA;AAED,mFAAmF;AACnF,eAAO,MAAM,MAAM,SAAU,YAAY,UAAU,MAAM,QAAQ,MAAM,KAAG,YAOzE,CAAA;AAED,eAAO,MAAM,YAAY,UAChB,SAAS,aACL,MAAM,gBACe;IAAE,OAAO,CAAC,EAAE,MAAM,CAAA;CAAE,KACnD,OAAO,CAAC,QAAQ,CA6ClB,CAAA"}

package/dist/embeddings/embed.js

@@ -0,0 +1,140 @@
+import { readFileSync } from "node:fs";
+import { availableParallelism } from "node:os";
+import { basename, join } from "node:path";
+import { filesPresent, sharedModelsDirectory } from "../speech/install.js";
+export const textModelsDirectory = (env = process.env) => sharedModelsDirectory("text", env);
+export const placedText = (model, directory) => model.files.map((file) => [file, join(directory, model.id, file.name)]);
+export const isTextModelInstalled = (model, directory) => filesPresent(placedText(model, directory));
+/** Past 8 threads one session gains nothing; Bun takes 1 unless told (`2026-10-02-embeddings.md`). */
+export const defaultThreads = () => Math.min(8, availableParallelism());
+export const WARM_IDLE_MS = 10 * 60_000;
+export const warmEmbedders = ({ idleMs = WARM_IDLE_MS } = {}) => {
+    const held = new Map();
+    const timers = new Map();
+    const running = new Map();
+    const drop = (key) => {
+        const one = held.get(key);
+        held.delete(key);
+        clearTimeout(timers.get(key));
+        timers.delete(key);
+        return one;
+    };
+    const rearm = (key) => {
+        clearTimeout(timers.get(key));
+        const timer = setTimeout(() => {
+            // A search still running rearms the clock when it ends.
+            if (running.get(key))
+                return;
+            drop(key)
+                ?.then((one) => one.close())
+                .catch(() => { });
+        }, idleMs);
+        // The MCP server exits when its client goes; a timer must not hold it open.
+        timer.unref?.();
+        timers.set(key, timer);
+    };
+    return {
+        get: (key, open) => {
+            let opening = held.get(key);
+            if (!opening) {
+                const started = open().then((one) => serial(one, {
+                    starts: () => running.set(key, (running.get(key) ?? 0) + 1),
+                    ends: () => {
+                        running.set(key, (running.get(key) ?? 1) - 1);
+                        rearm(key);
+                    },
+                }));
+                held.set(key, started);
+                // A model not downloaded yet must not stay refused once it is.
+                started.catch(() => {
+                    if (held.get(key) === started)
+                        drop(key);
+                });
+                opening = started;
+            }
+            rearm(key);
+            return opening;
+        },
+        close: async () => {
+            const all = [...held.keys()].map(drop);
+            await Promise.allSettled(all.map(async (one) => (await one)?.close()));
+        },
+    };
+};
+/** Two MCP calls may arrive at once; a session is not documented to take overlapping runs. */
+const serial = (embedder, { starts, ends }) => {
+    let last = Promise.resolve();
+    return {
+        model: embedder.model,
+        embed: (texts, kind) => {
+            starts();
+            const next = last.then(() => embedder.embed(texts, kind));
+            last = next.then(ends, ends);
+            return next;
+        },
+        close: async () => {
+            await last;
+            await embedder.close();
+        },
+    };
+};
+/** A text longer than the model's limit keeps its beginning and the closing special token. */
+export const truncated = (ids, max) => ids.length <= max ? ids : [...ids.slice(0, max - 1), ids.at(-1)];
+export const unit = (vector) => {
+    let sum = 0;
+    for (const value of vector)
+        sum += value * value;
+    const length = Math.sqrt(sum);
+    return length === 0 ? vector : vector.map((value) => value / length);
+};
+/** The mean over a single text's tokens: one text a run, so no padding to mask. */
+export const meanOf = (data, tokens, dims) => {
+    const mean = new Float32Array(dims);
+    for (let token = 0; token < tokens; token++) {
+        for (let dim = 0; dim < dims; dim++)
+            mean[dim] = mean[dim] + data[token * dims + dim] / tokens;
+    }
+    return mean;
+};
+export const openEmbedder = async (model, directory, { threads = defaultThreads() } = {}) => {
+    const [{ Tokenizer }, ort] = await Promise.all([
+        import("@huggingface/tokenizers"),
+        import("@wirecat/cli-messaging-onnx"),
+    ]);
+    const path = (name) => join(directory, model.id, name);
+    const tokenizer = new Tokenizer(JSON.parse(readFileSync(path("tokenizer.json"), "utf8")), JSON.parse(readFileSync(path("tokenizer_config.json"), "utf8")));
+    ort.env.wasm.numThreads = threads;
+    const session = await ort.InferenceSession.create(readFileSync(path(model.onnx)), model.externalData
+        ? { externalData: [{ path: basename(model.externalData), data: readFileSync(path(model.externalData)) }] }
+        : {});
+    const typeIds = session.inputNames.includes("token_type_ids");
+    const one = async (text) => {
+        const ids = truncated(tokenizer.encode(text).ids, model.maxTokens);
+        const shape = [1, ids.length];
+        const feeds = {
+            input_ids: new ort.Tensor("int64", BigInt64Array.from(ids, BigInt), shape),
+            attention_mask: new ort.Tensor("int64", new BigInt64Array(ids.length).fill(1n), shape),
+        };
+        if (typeIds)
+            feeds.token_type_ids = new ort.Tensor("int64", new BigInt64Array(ids.length), shape);
+        const output = await session.run(feeds);
+        if (model.pooling === "sentence_embedding") {
+            return unit(Float32Array.from(output.sentence_embedding?.data));
+        }
+        const hidden = output.last_hidden_state;
+        if (!hidden)
+            throw new Error(`${model.id} gave no last_hidden_state`);
+        return unit(meanOf(hidden.data, ids.length, hidden.dims.at(-1)));
+    };
+    return {
+        model,
+        embed: async (texts, kind) => {
+            const vectors = [];
+            for (const text of texts)
+                vectors.push(await one(`${model.prefix[kind]}${text}`));
+            return vectors;
+        },
+        close: () => session.release(),
+    };
+};
+//# sourceMappingURL=embed.js.map

package/dist/embeddings/embed.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"embed.js","sourceRoot":"","sources":["../../src/embeddings/embed.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,MAAM,SAAS,CAAA;AACtC,OAAO,EAAE,oBAAoB,EAAE,MAAM,SAAS,CAAA;AAC9C,OAAO,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,WAAW,CAAA;AAC1C,OAAO,EAAE,YAAY,EAAE,qBAAqB,EAAE,MAAM,sBAAsB,CAAA;AAG1E,MAAM,CAAC,MAAM,mBAAmB,GAAG,CAAC,GAAG,GAAsB,OAAO,CAAC,GAAG,EAAU,EAAE,CAAC,qBAAqB,CAAC,MAAM,EAAE,GAAG,CAAC,CAAA;AAEvH,MAAM,CAAC,MAAM,UAAU,GAAG,CAAC,KAAgB,EAAE,SAAiB,EAAE,EAAE,CAChE,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,IAAI,EAAE,IAAI,CAAC,SAAS,EAAE,KAAK,CAAC,EAAE,EAAE,IAAI,CAAC,IAAI,CAAC,CAA0B,CAAC,CAAA;AAElG,MAAM,CAAC,MAAM,oBAAoB,GAAG,CAAC,KAAgB,EAAE,SAAiB,EAAW,EAAE,CACnF,YAAY,CAAC,UAAU,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC,CAAA;AAE5C,sGAAsG;AACtG,MAAM,CAAC,MAAM,cAAc,GAAG,GAAW,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,oBAAoB,EAAE,CAAC,CAAA;AAsB/E,MAAM,CAAC,MAAM,YAAY,GAAG,EAAE,GAAG,MAAM,CAAA;AAEvC,MAAM,CAAC,MAAM,aAAa,GAAG,CAAC,EAAE,MAAM,GAAG,YAAY,EAAE,GAAwB,EAAE,EAAiB,EAAE;IAClG,MAAM,IAAI,GAAG,IAAI,GAAG,EAA6B,CAAA;IACjD,MAAM,MAAM,GAAG,IAAI,GAAG,EAAyC,CAAA;IAC/D,MAAM,OAAO,GAAG,IAAI,GAAG,EAAkB,CAAA;IACzC,MAAM,IAAI,GAAG,CAAC,GAAW,EAAE,EAAE;QAC3B,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,CAAA;QACzB,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,CAAA;QAChB,YAAY,CAAC,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAA;QAC7B,MAAM,CAAC,MAAM,CAAC,GAAG,CAAC,CAAA;QAClB,OAAO,GAAG,CAAA;IACZ,CAAC,CAAA;IACD,MAAM,KAAK,GAAG,CAAC,GAAW,EAAE,EAAE;QAC5B,YAAY,CAAC,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAA;QAC7B,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE;YAC5B,wDAAwD;YACxD,IAAI,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC;gBAAE,OAAM;YAC5B,IAAI,CAAC,GAAG,CAAC;gBACP,EAAE,IAAI,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,KAAK,EAAE,CAAC;iBAC3B,KAAK,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,CAAA;QACpB,CAAC,EAAE,MAAM,CAAC,CAAA;QACV,4EAA4E;QAC5E,KAAK,CAAC,KAAK,EAAE,EAAE,CAAA;QACf,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE,KAAK,CAAC,CAAA;IACxB,CAAC,CAAA;IACD,OAAO;QACL,GAAG,EAAE,CAAC,GAAG,EAAE,IAAI,EAAE,EAAE;YACjB,IAAI,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,CAAA;YAC3B,IAAI,CAAC,OAAO,EAAE,CAAC;gBACb,MAAM,OAAO,GAAG,IAAI,EAAE,CAAC,IAAI,CAAC,CAAC,GAAG,EAAE,EAAE,CAClC,MAAM,CAAC,GAAG,EAAE;oBACV,MAAM,EAAE,GAAG,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC;oBAC3D,IAAI,EAAE,GAAG,EAAE;wBACT,OAAO,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAA;wBAC7C,KAAK,CAAC,GAAG,CAAC,CAAA;oBACZ,CAAC;iBACF,CAAC,CACH,CAAA;gBACD,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,OAAO,CAAC,CAAA;gBACtB,+DAA+D;gBAC/D,OAAO,CAAC,KAAK,CAAC,GAAG,EAAE;oBACjB,IAAI,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,KAAK,OAAO;wBAAE,IAAI,CAAC,GAAG,CAAC,CAAA;gBAC1C,CAAC,CAAC,CAAA;gBACF,OAAO,GAAG,OAAO,CAAA;YACnB,CAAC;YACD,KAAK,CAAC,GAAG,CAAC,CAAA;YACV,OAAO,OAAO,CAAA;QAChB,CAAC;QACD,KAAK,EAAE,KAAK,IAAI,EAAE;YAChB,MAAM,GAAG,GAAG,CAAC,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,CAAA;YACtC,MAAM,OAAO,CAAC,UAAU,CAAC,GAAG,CAAC,GAAG,CAAC,KAAK,EAAE,GAAG,EAAE,EAAE,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,KAAK,EAAE,CAAC,CAAC,CAAA;QACxE,CAAC;KACF,CAAA;AACH,CAAC,CAAA;AAED,8FAA8F;AAC9F,MAAM,MAAM,GAAG,CAAC,QAAkB,EAAE,EAAE,MAAM,EAAE,IAAI,EAA4C,EAAY,EAAE;IAC1G,IAAI,IAAI,GAAqB,OAAO,CAAC,OAAO,EAAE,CAAA;IAC9C,OAAO;QACL,KAAK,EAAE,QAAQ,CAAC,KAAK;QACrB,KAAK,EAAE,CAAC,KAAK,EAAE,IAAI,EAAE,EAAE;YACrB,MAAM,EAAE,CAAA;YACR,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC,QAAQ,CAAC,KAAK,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC,CAAA;YACzD,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,IAAI,CAAC,CAAA;YAC5B,OAAO,IAAI,CAAA;QACb,CAAC;QACD,KAAK,EAAE,KAAK,IAAI,EAAE;YAChB,MAAM,IAAI,CAAA;YACV,MAAM,QAAQ,CAAC,KAAK,EAAE,CAAA;QACxB,CAAC;KACF,CAAA;AACH,CAAC,CAAA;AAED,8FAA8F;AAC9F,MAAM,CAAC,MAAM,SAAS,GAAG,CAAC,GAAa,EAAE,GAAW,EAAY,EAAE,CAChE,GAAG,CAAC,MAAM,IAAI,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,GAAG,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,GAAG,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC,CAAW,CAAC,CAAA;AAE5E,MAAM,CAAC,MAAM,IAAI,GAAG,CAAC,MAAoB,EAAgB,EAAE;IACzD,IAAI,GAAG,GAAG,CAAC,CAAA;IACX,KAAK,MAAM,KAAK,IAAI,MAAM;QAAE,GAAG,IAAI,KAAK,GAAG,KAAK,CAAA;IAChD,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAA;IAC7B,OAAO,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,GAAG,MAAM,CAAC,CAAA;AACtE,CAAC,CAAA;AAED,mFAAmF;AACnF,MAAM,CAAC,MAAM,MAAM,GAAG,CAAC,IAAkB,EAAE,MAAc,EAAE,IAAY,EAAgB,EAAE;IACvF,MAAM,IAAI,GAAG,IAAI,YAAY,CAAC,IAAI,CAAC,CAAA;IACnC,KAAK,IAAI,KAAK,GAAG,CAAC,EAAE,KAAK,GAAG,MAAM,EAAE,KAAK,EAAE,EAAE,CAAC;QAC5C,KAAK,IAAI,GAAG,GAAG,CAAC,EAAE,GAAG,GAAG,IAAI,EAAE,GAAG,EAAE;YACjC,IAAI,CAAC,GAAG,CAAC,GAAI,IAAI,CAAC,GAAG,CAAY,GAAI,IAAI,CAAC,KAAK,GAAG,IAAI,GAAG,GAAG,CAAY,GAAG,MAAM,CAAA;IACrF,CAAC;IACD,OAAO,IAAI,CAAA;AACb,CAAC,CAAA;AAED,MAAM,CAAC,MAAM,YAAY,GAAG,KAAK,EAC/B,KAAgB,EAChB,SAAiB,EACjB,EAAE,OAAO,GAAG,cAAc,EAAE,EAAE,GAAyB,EAAE,EACtC,EAAE;IACrB,MAAM,CAAC,EAAE,SAAS,EAAE,EAAE,GAAG,CAAC,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;QAC7C,MAAM,CAAC,yBAAyB,CAAC;QACjC,MAAM,CAAC,6BAA6B,CAAC;KACtC,CAAC,CAAA;IACF,MAAM,IAAI,GAAG,CAAC,IAAY,EAAE,EAAE,CAAC,IAAI,CAAC,SAAS,EAAE,KAAK,CAAC,EAAE,EAAE,IAAI,CAAC,CAAA;IAC9D,MAAM,SAAS,GAAG,IAAI,SAAS,CAC7B,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,CAAC,gBAAgB,CAAC,EAAE,MAAM,CAAC,CAAC,EACxD,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,CAAC,uBAAuB,CAAC,EAAE,MAAM,CAAC,CAAC,CAChE,CAAA;IACD,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,UAAU,GAAG,OAAO,CAAA;IACjC,MAAM,OAAO,GAAG,MAAM,GAAG,CAAC,gBAAgB,CAAC,MAAM,CAC/C,YAAY,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,EAC9B,KAAK,CAAC,YAAY;QAChB,CAAC,CAAC,EAAE,YAAY,EAAE,CAAC,EAAE,IAAI,EAAE,QAAQ,CAAC,KAAK,CAAC,YAAY,CAAC,EAAE,IAAI,EAAE,YAAY,CAAC,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,CAAC,EAAE,CAAC,EAAE;QAC1G,CAAC,CAAC,EAAE,CACP,CAAA;IACD,MAAM,OAAO,GAAG,OAAO,CAAC,UAAU,CAAC,QAAQ,CAAC,gBAAgB,CAAC,CAAA;IAE7D,MAAM,GAAG,GAAG,KAAK,EAAE,IAAY,EAAyB,EAAE;QACxD,MAAM,GAAG,GAAG,SAAS,CAAC,SAAS,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,GAAG,EAAE,KAAK,CAAC,SAAS,CAAC,CAAA;QAClE,MAAM,KAAK,GAAG,CAAC,CAAC,EAAE,GAAG,CAAC,MAAM,CAAC,CAAA;QAC7B,MAAM,KAAK,GAAoD;YAC7D,SAAS,EAAE,IAAI,GAAG,CAAC,MAAM,CAAC,OAAO,EAAE,aAAa,CAAC,IAAI,CAAC,GAAG,EAAE,MAAM,CAAC,EAAE,KAAK,CAAC;YAC1E,cAAc,EAAE,IAAI,GAAG,CAAC,MAAM,CAAC,OAAO,EAAE,IAAI,aAAa,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,KAAK,CAAC;SACvF,CAAA;QACD,IAAI,OAAO;YAAE,KAAK,CAAC,cAAc,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,OAAO,EAAE,IAAI,aAAa,CAAC,GAAG,CAAC,MAAM,CAAC,EAAE,KAAK,CAAC,CAAA;QACjG,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,CAAA;QACvC,IAAI,KAAK,CAAC,OAAO,KAAK,oBAAoB,EAAE,CAAC;YAC3C,OAAO,IAAI,CAAC,YAAY,CAAC,IAAI,CAAC,MAAM,CAAC,kBAAkB,EAAE,IAAoB,CAAC,CAAC,CAAA;QACjF,CAAC;QACD,MAAM,MAAM,GAAG,MAAM,CAAC,iBAAiB,CAAA;QACvC,IAAI,CAAC,MAAM;YAAE,MAAM,IAAI,KAAK,CAAC,GAAG,KAAK,CAAC,EAAE,4BAA4B,CAAC,CAAA;QACrE,OAAO,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,IAAoB,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,CAAW,CAAC,CAAC,CAAA;IAC5F,CAAC,CAAA;IAED,OAAO;QACL,KAAK;QACL,KAAK,EAAE,KAAK,EAAE,KAAK,EAAE,IAAI,EAAE,EAAE;YAC3B,MAAM,OAAO,GAAmB,EAAE,CAAA;YAClC,KAAK,MAAM,IAAI,IAAI,KAAK;gBAAE,OAAO,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,GAAG,KAAK,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,IAAI,EAAE,CAAC,CAAC,CAAA;YACjF,OAAO,OAAO,CAAA;QAChB,CAAC;QACD,KAAK,EAAE,GAAG,EAAE,CAAC,OAAO,CAAC,OAAO,EAAE;KAC/B,CAAA;AACH,CAAC,CAAA"}

package/dist/embeddings/models.d.ts

@@ -0,0 +1,37 @@
+import type { ModelFile } from "../speech/models.js";
+/**
+ * The text embedding models `conversations embed` can use, each **pinned to one commit** and checked by
+ * sha256, as the speech models are. Measured 2026-10-02 through `@wirecat/cli-messaging-onnx`, 4 threads, a
+ * Ryzen AI 9 HX 470 laptop, 300-token chunks, Node 24 and Bun 1.3.14 alike
+ * (`docs/storage/research/2026-10-02-embeddings.md`): e5-small ~10 chunks a second, EmbeddingGemma ~1.5.
+ */
+export interface TextModel {
+    id: string;
+    title: string;
+    languages: string;
+    licence: string;
+    /** A licence the user accepts before the model is downloaded, or nothing for an open one. */
+    terms?: string;
+    dims: number;
+    /** What the model was trained on; a longer text is cut. */
+    maxTokens: number;
+    /** About how many ~300-token chunks a second one session embeds on 8 threads, for the estimate. */
+    chunksPerSecond: number;
+    /** How the model's output becomes one vector. */
+    pooling: "mean" | "sentence_embedding";
+    /** Prepended to a search and to a stored chunk: the model was trained with them. */
+    prefix: {
+        query: string;
+        passage: string;
+    };
+    files: ModelFile[];
+    /** The ONNX file among `files`, and the weights it keeps beside itself, if any. */
+    onnx: string;
+    externalData?: string;
+}
+export declare const TEXT_MODELS: TextModel[];
+export declare const DEFAULT_TEXT_MODEL = "e5-small";
+export declare const textModel: (id: string) => TextModel;
+/** The cosine a meaning hit must beat, measured in bench/search-quality; an unmeasured model keeps every hit. */
+export declare const meaningFloor: (key: string) => number;
+//# sourceMappingURL=models.d.ts.map

package/dist/embeddings/models.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"models.d.ts","sourceRoot":"","sources":["../../src/embeddings/models.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,qBAAqB,CAAA;AAEpD;;;;;GAKG;AACH,MAAM,WAAW,SAAS;IACxB,EAAE,EAAE,MAAM,CAAA;IACV,KAAK,EAAE,MAAM,CAAA;IACb,SAAS,EAAE,MAAM,CAAA;IACjB,OAAO,EAAE,MAAM,CAAA;IACf,6FAA6F;IAC7F,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,IAAI,EAAE,MAAM,CAAA;IACZ,2DAA2D;IAC3D,SAAS,EAAE,MAAM,CAAA;IACjB,mGAAmG;IACnG,eAAe,EAAE,MAAM,CAAA;IACvB,iDAAiD;IACjD,OAAO,EAAE,MAAM,GAAG,oBAAoB,CAAA;IACtC,oFAAoF;IACpF,MAAM,EAAE;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAA;KAAE,CAAA;IAC1C,KAAK,EAAE,SAAS,EAAE,CAAA;IAClB,mFAAmF;IACnF,IAAI,EAAE,MAAM,CAAA;IACZ,YAAY,CAAC,EAAE,MAAM,CAAA;CACtB;AAQD,eAAO,MAAM,WAAW,EAAE,SAAS,EA0ElC,CAAA;AAED,eAAO,MAAM,kBAAkB,aAAa,CAAA;AAE5C,eAAO,MAAM,SAAS,OAAQ,MAAM,KAAG,SAStC,CAAA;AAED,iHAAiH;AACjH,eAAO,MAAM,YAAY,QAAS,MAAM,KAAG,MAAkD,CAAA"}

package/dist/embeddings/models.js

@@ -0,0 +1,90 @@
+import { CliError } from "@wirecat/cli-base";
+const huggingFace = (repository, commit) => (name) => `https://huggingface.co/${repository}/resolve/${commit}/${name}`;
+const e5 = huggingFace("Xenova/multilingual-e5-small", "761b726dd34fb83930e26aab4e9ac3899aa1fa78");
+const gemma = huggingFace("onnx-community/embeddinggemma-300m-ONNX", "5090578d9565bb06545b4552f76e6bc2c93e4a66");
+export const TEXT_MODELS = [
+    {
+        id: "e5-small",
+        title: "multilingual-e5-small, 8-bit",
+        languages: "about 100 languages, Russian and English among them",
+        licence: "MIT",
+        dims: 384,
+        maxTokens: 512,
+        // 100k-message corpus, one 8-thread session on Ryzen AI 9 HX 470: 31–32 chunks/s.
+        chunksPerSecond: 31,
+        pooling: "mean",
+        prefix: { query: "query: ", passage: "passage: " },
+        onnx: "onnx/model_quantized.onnx",
+        files: [
+            {
+                name: "onnx/model_quantized.onnx",
+                url: e5("onnx/model_quantized.onnx"),
+                sha256: "f80102d3f2a1229f387d3c81909990d8945513e347b0eab049f7de3c6f98c193",
+                bytes: 118_308_185,
+            },
+            {
+                name: "tokenizer.json",
+                url: e5("tokenizer.json"),
+                sha256: "0b44a9d7b51c3c62626640cda0e2c2f70fdacdc25bbbd68038369d14ebdf4c39",
+                bytes: 17_082_730,
+            },
+            {
+                name: "tokenizer_config.json",
+                url: e5("tokenizer_config.json"),
+                sha256: "a1d6bc8734a6f635dc158508bef000f8e2e5a759c7d92f984b2c86e5ff53425b",
+                bytes: 443,
+            },
+        ],
+    },
+    {
+        id: "embeddinggemma",
+        title: "EmbeddingGemma 300M, 4-bit",
+        languages: "over 100 languages — the best of the two on chat, about seven times slower",
+        licence: "Gemma Terms of Use",
+        terms: "https://ai.google.dev/gemma/terms",
+        dims: 768,
+        maxTokens: 2048,
+        chunksPerSecond: 2,
+        pooling: "sentence_embedding",
+        prefix: { query: "task: search result | query: ", passage: "title: none | text: " },
+        onnx: "onnx/model_q4.onnx",
+        externalData: "onnx/model_q4.onnx_data",
+        files: [
+            {
+                name: "onnx/model_q4.onnx",
+                url: gemma("onnx/model_q4.onnx"),
+                sha256: "ad1dfee81a70f7944b9b9d1cc6e48075b832881cf33fab2f2b248be78f3f0043",
+                bytes: 519_322,
+            },
+            {
+                name: "onnx/model_q4.onnx_data",
+                url: gemma("onnx/model_q4.onnx_data"),
+                sha256: "599962c3143b040de2dd05e5975be3e9091dd067cacc6a8f7186e3203bab9e02",
+                bytes: 196_725_760,
+            },
+            {
+                name: "tokenizer.json",
+                url: gemma("tokenizer.json"),
+                sha256: "4dda02faaf32bc91031dc8c88457ac272b00c1016cc679757d1c441b248b9c47",
+                bytes: 20_323_312,
+            },
+            {
+                name: "tokenizer_config.json",
+                url: gemma("tokenizer_config.json"),
+                sha256: "3ca953eea6c3c9fcda9cf3df22949ff18b216f7c74bd6459230f3f1013953f3a",
+                bytes: 1_156_830,
+            },
+        ],
+    },
+];
+export const DEFAULT_TEXT_MODEL = "e5-small";
+export const textModel = (id) => {
+    const found = TEXT_MODELS.find((model) => model.id === id);
+    if (!found) {
+        throw new CliError("validation_error", `no text model ${id} — one of: ${TEXT_MODELS.map((model) => model.id).join(", ")}`);
+    }
+    return found;
+};
+/** The cosine a meaning hit must beat, measured in bench/search-quality; an unmeasured model keeps every hit. */
+export const meaningFloor = (key) => (key === "local:e5-small:384" ? 0.8 : 0);
+//# sourceMappingURL=models.js.map

package/dist/embeddings/models.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"models.js","sourceRoot":"","sources":["../../src/embeddings/models.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,mBAAmB,CAAA;AA+B5C,MAAM,WAAW,GAAG,CAAC,UAAkB,EAAE,MAAc,EAAE,EAAE,CAAC,CAAC,IAAY,EAAE,EAAE,CAC3E,0BAA0B,UAAU,YAAY,MAAM,IAAI,IAAI,EAAE,CAAA;AAElE,MAAM,EAAE,GAAG,WAAW,CAAC,8BAA8B,EAAE,0CAA0C,CAAC,CAAA;AAClG,MAAM,KAAK,GAAG,WAAW,CAAC,yCAAyC,EAAE,0CAA0C,CAAC,CAAA;AAEhH,MAAM,CAAC,MAAM,WAAW,GAAgB;IACtC;QACE,EAAE,EAAE,UAAU;QACd,KAAK,EAAE,8BAA8B;QACrC,SAAS,EAAE,qDAAqD;QAChE,OAAO,EAAE,KAAK;QACd,IAAI,EAAE,GAAG;QACT,SAAS,EAAE,GAAG;QACd,kFAAkF;QAClF,eAAe,EAAE,EAAE;QACnB,OAAO,EAAE,MAAM;QACf,MAAM,EAAE,EAAE,KAAK,EAAE,SAAS,EAAE,OAAO,EAAE,WAAW,EAAE;QAClD,IAAI,EAAE,2BAA2B;QACjC,KAAK,EAAE;YACL;gBACE,IAAI,EAAE,2BAA2B;gBACjC,GAAG,EAAE,EAAE,CAAC,2BAA2B,CAAC;gBACpC,MAAM,EAAE,kEAAkE;gBAC1E,KAAK,EAAE,WAAW;aACnB;YACD;gBACE,IAAI,EAAE,gBAAgB;gBACtB,GAAG,EAAE,EAAE,CAAC,gBAAgB,CAAC;gBACzB,MAAM,EAAE,kEAAkE;gBAC1E,KAAK,EAAE,UAAU;aAClB;YACD;gBACE,IAAI,EAAE,uBAAuB;gBAC7B,GAAG,EAAE,EAAE,CAAC,uBAAuB,CAAC;gBAChC,MAAM,EAAE,kEAAkE;gBAC1E,KAAK,EAAE,GAAG;aACX;SACF;KACF;IACD;QACE,EAAE,EAAE,gBAAgB;QACpB,KAAK,EAAE,4BAA4B;QACnC,SAAS,EAAE,4EAA4E;QACvF,OAAO,EAAE,oBAAoB;QAC7B,KAAK,EAAE,mCAAmC;QAC1C,IAAI,EAAE,GAAG;QACT,SAAS,EAAE,IAAI;QACf,eAAe,EAAE,CAAC;QAClB,OAAO,EAAE,oBAAoB;QAC7B,MAAM,EAAE,EAAE,KAAK,EAAE,+BAA+B,EAAE,OAAO,EAAE,sBAAsB,EAAE;QACnF,IAAI,EAAE,oBAAoB;QAC1B,YAAY,EAAE,yBAAyB;QACvC,KAAK,EAAE;YACL;gBACE,IAAI,EAAE,oBAAoB;gBAC1B,GAAG,EAAE,KAAK,CAAC,oBAAoB,CAAC;gBAChC,MAAM,EAAE,kEAAkE;gBAC1E,KAAK,EAAE,OAAO;aACf;YACD;gBACE,IAAI,EAAE,yBAAyB;gBAC/B,GAAG,EAAE,KAAK,CAAC,yBAAyB,CAAC;gBACrC,MAAM,EAAE,kEAAkE;gBAC1E,KAAK,EAAE,WAAW;aACnB;YACD;gBACE,IAAI,EAAE,gBAAgB;gBACtB,GAAG,EAAE,KAAK,CAAC,gBAAgB,CAAC;gBAC5B,MAAM,EAAE,kEAAkE;gBAC1E,KAAK,EAAE,UAAU;aAClB;YACD;gBACE,IAAI,EAAE,uBAAuB;gBAC7B,GAAG,EAAE,KAAK,CAAC,uBAAuB,CAAC;gBACnC,MAAM,EAAE,kEAAkE;gBAC1E,KAAK,EAAE,SAAS;aACjB;SACF;KACF;CACF,CAAA;AAED,MAAM,CAAC,MAAM,kBAAkB,GAAG,UAAU,CAAA;AAE5C,MAAM,CAAC,MAAM,SAAS,GAAG,CAAC,EAAU,EAAa,EAAE;IACjD,MAAM,KAAK,GAAG,WAAW,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,EAAE,KAAK,EAAE,CAAC,CAAA;IAC1D,IAAI,CAAC,KAAK,EAAE,CAAC;QACX,MAAM,IAAI,QAAQ,CAChB,kBAAkB,EAClB,iBAAiB,EAAE,cAAc,WAAW,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CACnF,CAAA;IACH,CAAC;IACD,OAAO,KAAK,CAAA;AACd,CAAC,CAAA;AAED,iHAAiH;AACjH,MAAM,CAAC,MAAM,YAAY,GAAG,CAAC,GAAW,EAAU,EAAE,CAAC,CAAC,GAAG,KAAK,oBAAoB,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAA"}

package/dist/embeddings/pool.d.ts

@@ -0,0 +1,17 @@
+import { type Embedder } from "./embed.js";
+import type { TextModel } from "./models.js";
+/** Each worker holds its own copy of the model: 550–650 MB measured for e5-small (phase 5 E12). */
+export declare const WORKER_BYTES = 700000000;
+/**
+ * One session in this thread, or `workers` sessions in worker threads with the threads split between
+ * them. Correction 2026-10-03: the 100k-message benchmark measured 3 workers at 1.04–1.1× one
+ * session on 8 threads, with ~2.6–2.7 GB RSS against Node's 1.2 GB (bench/embeddings/README.md).
+ * The research's ~1.8× used 4 threads per worker against one session on 4, not this allocation.
+ * Refuses a count whose copies would not fit in free memory.
+ */
+export declare const openPool: (model: TextModel, directory: string, { workers, threads, free, }?: {
+    workers?: number;
+    threads?: number;
+    free?: number;
+}) => Promise<Embedder>;
+//# sourceMappingURL=pool.d.ts.map

package/dist/embeddings/pool.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"pool.d.ts","sourceRoot":"","sources":["../../src/embeddings/pool.ts"],"names":[],"mappings":"AAEA,OAAO,EAAkB,KAAK,QAAQ,EAAgB,MAAM,YAAY,CAAA;AACxE,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,aAAa,CAAA;AAE5C,mGAAmG;AACnG,eAAO,MAAM,YAAY,YAAc,CAAA;AAEvC;;;;;;GAMG;AACH,eAAO,MAAM,QAAQ,UACZ,SAAS,aACL,MAAM,gCAKd;IAAE,OAAO,CAAC,EAAE,MAAM,CAAC;IAAC,OAAO,CAAC,EAAE,MAAM,CAAC;IAAC,IAAI,CAAC,EAAE,MAAM,CAAA;CAAE,KACvD,OAAO,CAAC,QAAQ,CAUlB,CAAA"}

package/dist/embeddings/pool.js

@@ -0,0 +1,22 @@
+import { freemem } from "node:os";
+import { CliError } from "@wirecat/cli-base";
+import { defaultThreads, openEmbedder } from "./embed.js";
+/** Each worker holds its own copy of the model: 550–650 MB measured for e5-small (phase 5 E12). */
+export const WORKER_BYTES = 700_000_000;
+/**
+ * One session in this thread, or `workers` sessions in worker threads with the threads split between
+ * them. Correction 2026-10-03: the 100k-message benchmark measured 3 workers at 1.04–1.1× one
+ * session on 8 threads, with ~2.6–2.7 GB RSS against Node's 1.2 GB (bench/embeddings/README.md).
+ * The research's ~1.8× used 4 threads per worker against one session on 4, not this allocation.
+ * Refuses a count whose copies would not fit in free memory.
+ */
+export const openPool = async (model, directory, { workers = 1, threads = defaultThreads(), free = freemem(), } = {}) => {
+    if (workers <= 1)
+        return openEmbedder(model, directory, { threads });
+    if (workers * WORKER_BYTES > free) {
+        throw new CliError("validation_error", `${workers} workers need about ${Math.round((workers * WORKER_BYTES) / 1e9)} GB, and ${Math.floor(free / 1e9)} GB is free — use fewer`);
+    }
+    const { openWorkers } = await import("./workers.js");
+    return openWorkers(model, directory, { workers, threads: Math.max(1, Math.floor(threads / workers)) });
+};
+//# sourceMappingURL=pool.js.map

package/dist/embeddings/pool.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"pool.js","sourceRoot":"","sources":["../../src/embeddings/pool.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,SAAS,CAAA;AACjC,OAAO,EAAE,QAAQ,EAAE,MAAM,mBAAmB,CAAA;AAC5C,OAAO,EAAE,cAAc,EAAiB,YAAY,EAAE,MAAM,YAAY,CAAA;AAGxE,mGAAmG;AACnG,MAAM,CAAC,MAAM,YAAY,GAAG,WAAW,CAAA;AAEvC;;;;;;GAMG;AACH,MAAM,CAAC,MAAM,QAAQ,GAAG,KAAK,EAC3B,KAAgB,EAChB,SAAiB,EACjB,EACE,OAAO,GAAG,CAAC,EACX,OAAO,GAAG,cAAc,EAAE,EAC1B,IAAI,GAAG,OAAO,EAAE,GACjB,GAA0D,EAAE,EAC1C,EAAE;IACrB,IAAI,OAAO,IAAI,CAAC;QAAE,OAAO,YAAY,CAAC,KAAK,EAAE,SAAS,EAAE,EAAE,OAAO,EAAE,CAAC,CAAA;IACpE,IAAI,OAAO,GAAG,YAAY,GAAG,IAAI,EAAE,CAAC;QAClC,MAAM,IAAI,QAAQ,CAChB,kBAAkB,EAClB,GAAG,OAAO,uBAAuB,IAAI,CAAC,KAAK,CAAC,CAAC,OAAO,GAAG,YAAY,CAAC,GAAG,GAAG,CAAC,YAAY,IAAI,CAAC,KAAK,CAAC,IAAI,GAAG,GAAG,CAAC,yBAAyB,CACvI,CAAA;IACH,CAAC;IACD,MAAM,EAAE,WAAW,EAAE,GAAG,MAAM,MAAM,CAAC,cAAc,CAAC,CAAA;IACpD,OAAO,WAAW,CAAC,KAAK,EAAE,SAAS,EAAE,EAAE,OAAO,EAAE,OAAO,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,OAAO,GAAG,OAAO,CAAC,CAAC,EAAE,CAAC,CAAA;AACxG,CAAC,CAAA"}

package/dist/embeddings/process.d.ts

@@ -0,0 +1,11 @@
+import type { Embedder } from "./embed.js";
+import type { TextModel } from "./models.js";
+/**
+ * One session in a process of its own, run by the same runtime. When it exits, all of its memory goes back:
+ * closing the session gives back ~0.1 of ~1 GB, and on Bun ending a worker thread gives back none for good
+ * (measured 2026-10-02, docs/storage/search-indexes.md).
+ */
+export declare const openProcess: (model: TextModel, directory: string, { threads }: {
+    threads: number;
+}) => Promise<Embedder>;
+//# sourceMappingURL=process.d.ts.map

package/dist/embeddings/process.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"process.d.ts","sourceRoot":"","sources":["../../src/embeddings/process.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,YAAY,CAAA;AAC1C,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,aAAa,CAAA;AAI5C;;;;GAIG;AACH,eAAO,MAAM,WAAW,UACf,SAAS,aACL,MAAM,eACJ;IAAE,OAAO,EAAE,MAAM,CAAA;CAAE,KAC/B,OAAO,CAAC,QAAQ,CAwDlB,CAAA"}

package/dist/embeddings/process.js

@@ -0,0 +1,67 @@
+import { spawn } from "node:child_process";
+import { createInterface } from "node:readline";
+import { fileURLToPath } from "node:url";
+/**
+ * One session in a process of its own, run by the same runtime. When it exits, all of its memory goes back:
+ * closing the session gives back ~0.1 of ~1 GB, and on Bun ending a worker thread gives back none for good
+ * (measured 2026-10-02, docs/storage/search-indexes.md).
+ */
+export const openProcess = async (model, directory, { threads }) => {
+    const child = spawn(process.execPath, [fileURLToPath(new URL("./child.js", import.meta.url)), JSON.stringify({ model, directory, threads })], { stdio: ["pipe", "pipe", "pipe"] });
+    const pending = new Map();
+    let said = "";
+    child.stderr.setEncoding("utf8").on("data", (text) => {
+        said = (said + text).slice(-2_000);
+    });
+    let ready = () => { };
+    let refused = (_error) => { };
+    const started = new Promise((resolve, reject) => {
+        ready = resolve;
+        refused = reject;
+    });
+    let gone;
+    const stopped = (error) => {
+        gone ??= error;
+        refused(error);
+        for (const job of pending.values())
+            job.reject(error);
+        pending.clear();
+    };
+    const exited = new Promise((resolve) => {
+        child.once("exit", (code) => {
+            const last = said.trim().split("\n").at(-1);
+            stopped(new Error(`the embedding process stopped (exit ${code})${last ? `: ${last}` : ""}`));
+            resolve();
+        });
+    });
+    child.once("error", stopped);
+    createInterface({ input: child.stdout }).on("line", (line) => {
+        const answer = JSON.parse(line);
+        if ("ready" in answer)
+            return ready();
+        const job = pending.get(answer.id);
+        pending.delete(answer.id);
+        if (answer.error !== undefined || !answer.vectors)
+            job?.reject(new Error(answer.error ?? "no vectors"));
+        else
+            job?.resolve(answer.vectors.map((vector) => Float32Array.from(vector)));
+    });
+    await started;
+    let next = 0;
+    return {
+        model,
+        embed: (texts, kind) => new Promise((resolve, reject) => {
+            if (gone)
+                return reject(gone);
+            const id = next++;
+            pending.set(id, { resolve, reject });
+            child.stdin.write(`${JSON.stringify({ id, texts, kind })}\n`);
+        }),
+        close: async () => {
+            if (child.exitCode === null && child.signalCode === null)
+                child.stdin.end();
+            await exited;
+        },
+    };
+};
+//# sourceMappingURL=process.js.map

package/dist/embeddings/process.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"process.js","sourceRoot":"","sources":["../../src/embeddings/process.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,EAAE,MAAM,oBAAoB,CAAA;AAC1C,OAAO,EAAE,eAAe,EAAE,MAAM,eAAe,CAAA;AAC/C,OAAO,EAAE,aAAa,EAAE,MAAM,UAAU,CAAA;AAMxC;;;;GAIG;AACH,MAAM,CAAC,MAAM,WAAW,GAAG,KAAK,EAC9B,KAAgB,EAChB,SAAiB,EACjB,EAAE,OAAO,EAAuB,EACb,EAAE;IACrB,MAAM,KAAK,GAAG,KAAK,CACjB,OAAO,CAAC,QAAQ,EAChB,CAAC,aAAa,CAAC,IAAI,GAAG,CAAC,YAAY,EAAE,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,SAAS,EAAE,OAAO,EAAE,CAAC,CAAC,EACtG,EAAE,KAAK,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,CACpC,CAAA;IACD,MAAM,OAAO,GAAG,IAAI,GAAG,EAA0F,CAAA;IACjH,IAAI,IAAI,GAAG,EAAE,CAAA;IACb,KAAK,CAAC,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,IAAY,EAAE,EAAE;QAC3D,IAAI,GAAG,CAAC,IAAI,GAAG,IAAI,CAAC,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAA;IACpC,CAAC,CAAC,CAAA;IACF,IAAI,KAAK,GAAG,GAAG,EAAE,GAAE,CAAC,CAAA;IACpB,IAAI,OAAO,GAAG,CAAC,MAAa,EAAE,EAAE,GAAE,CAAC,CAAA;IACnC,MAAM,OAAO,GAAG,IAAI,OAAO,CAAO,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;QACpD,KAAK,GAAG,OAAO,CAAA;QACf,OAAO,GAAG,MAAM,CAAA;IAClB,CAAC,CAAC,CAAA;IACF,IAAI,IAAuB,CAAA;IAC3B,MAAM,OAAO,GAAG,CAAC,KAAY,EAAE,EAAE;QAC/B,IAAI,KAAK,KAAK,CAAA;QACd,OAAO,CAAC,KAAK,CAAC,CAAA;QACd,KAAK,MAAM,GAAG,IAAI,OAAO,CAAC,MAAM,EAAE;YAAE,GAAG,CAAC,MAAM,CAAC,KAAK,CAAC,CAAA;QACrD,OAAO,CAAC,KAAK,EAAE,CAAA;IACjB,CAAC,CAAA;IACD,MAAM,MAAM,GAAG,IAAI,OAAO,CAAO,CAAC,OAAO,EAAE,EAAE;QAC3C,KAAK,CAAC,IAAI,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,EAAE;YAC1B,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAA;YAC3C,OAAO,CAAC,IAAI,KAAK,CAAC,uCAAuC,IAAI,IAAI,IAAI,CAAC,CAAC,CAAC,KAAK,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAA;YAC5F,OAAO,EAAE,CAAA;QACX,CAAC,CAAC,CAAA;IACJ,CAAC,CAAC,CAAA;IACF,KAAK,CAAC,IAAI,CAAC,OAAO,EAAE,OAAO,CAAC,CAAA;IAC5B,eAAe,CAAC,EAAE,KAAK,EAAE,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,EAAE;QAC3D,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAW,CAAA;QACzC,IAAI,OAAO,IAAI,MAAM;YAAE,OAAO,KAAK,EAAE,CAAA;QACrC,MAAM,GAAG,GAAG,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC,CAAA;QAClC,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC,CAAA;QACzB,IAAI,MAAM,CAAC,KAAK,KAAK,SAAS,IAAI,CAAC,MAAM,CAAC,OAAO;YAAE,GAAG,EAAE,MAAM,CAAC,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,IAAI,YAAY,CAAC,CAAC,CAAA;;YAClG,GAAG,EAAE,OAAO,CAAC,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,YAAY,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,CAAA;IAC9E,CAAC,CAAC,CAAA;IACF,MAAM,OAAO,CAAA;IACb,IAAI,IAAI,GAAG,CAAC,CAAA;IACZ,OAAO;QACL,KAAK;QACL,KAAK,EAAE,CAAC,KAAK,EAAE,IAAI,EAAE,EAAE,CACrB,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;YAC9B,IAAI,IAAI;gBAAE,OAAO,MAAM,CAAC,IAAI,CAAC,CAAA;YAC7B,MAAM,EAAE,GAAG,IAAI,EAAE,CAAA;YACjB,OAAO,CAAC,GAAG,CAAC,EAAE,EAAE,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAA;YACpC,KAAK,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,IAAI,CAAC,SAAS,CAAC,EAAE,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,IAAI,CAAC,CAAA;QAC/D,CAAC,CAAC;QACJ,KAAK,EAAE,KAAK,IAAI,EAAE;YAChB,IAAI,KAAK,CAAC,QAAQ,KAAK,IAAI,IAAI,KAAK,CAAC,UAAU,KAAK,IAAI;gBAAE,KAAK,CAAC,KAAK,CAAC,GAAG,EAAE,CAAA;YAC3E,MAAM,MAAM,CAAA;QACd,CAAC;KACF,CAAA;AACH,CAAC,CAAA"}

package/dist/embeddings/remote.d.ts

@@ -0,0 +1,43 @@
+import { type Embedder } from "./embed.js";
+/**
+ * An embedding model behind an OpenAI-shaped `/v1/embeddings` (phase 5 E11): OpenAI itself, or what
+ * serves the same request — Gemini's compatibility URL, Jina, Ollama and LM Studio on this machine.
+ * Facts: `docs/storage/research/2026-10-02-embedding-apis.md`.
+ */
+export interface RemoteModel {
+    provider: string;
+    model: string;
+    dims: number;
+    baseUrl: string;
+    /** Sent as `dimensions` when it differs from the model's own size. */
+    shorten: boolean;
+    /** USD per 1M input tokens, and the day the price page was read; nothing for a server of the user's own. */
+    price?: {
+        perMillion: number;
+        read: string;
+    };
+}
+export declare const OPENAI_URL = "https://api.openai.com/v1";
+/** OpenAI's models, from its pricing page as read on 2026-10-02. */
+export declare const OPENAI_MODELS: Record<string, {
+    dims: number;
+    perMillion: number;
+}>;
+export declare const DEFAULT_OPENAI_MODEL = "text-embedding-3-small";
+/** A server on this machine: nothing leaves it, so nothing is asked. */
+export declare const isLocal: (baseUrl: string) => boolean;
+export declare const remoteModel: ({ model, baseUrl, dims, }: {
+    model?: string;
+    baseUrl?: string;
+    dims?: number;
+}) => RemoteModel;
+/** The vector key of a remote model, beside `local:<model>:<dims>`. */
+export declare const remoteKey: ({ provider, model, dims }: RemoteModel) => string;
+export type Fetch = (url: string, init: RequestInit) => Promise<Response>;
+/** Up to `concurrency` requests at once; a 429 waits for `Retry-After`, a server error is tried again. */
+export declare const openRemote: (remote: RemoteModel, apiKey: string | undefined, { concurrency, fetch: post, tries }?: {
+    concurrency?: number;
+    fetch?: Fetch;
+    tries?: number;
+}) => Embedder;
+//# sourceMappingURL=remote.d.ts.map

package/dist/embeddings/remote.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"remote.d.ts","sourceRoot":"","sources":["../../src/embeddings/remote.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,KAAK,QAAQ,EAAQ,MAAM,YAAY,CAAA;AAGhD;;;;GAIG;AACH,MAAM,WAAW,WAAW;IAC1B,QAAQ,EAAE,MAAM,CAAA;IAChB,KAAK,EAAE,MAAM,CAAA;IACb,IAAI,EAAE,MAAM,CAAA;IACZ,OAAO,EAAE,MAAM,CAAA;IACf,sEAAsE;IACtE,OAAO,EAAE,OAAO,CAAA;IAChB,4GAA4G;IAC5G,KAAK,CAAC,EAAE;QAAE,UAAU,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAA;CAC7C;AAED,eAAO,MAAM,UAAU,8BAA8B,CAAA;AAErD,oEAAoE;AACpE,eAAO,MAAM,aAAa,EAAE,MAAM,CAAC,MAAM,EAAE;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,UAAU,EAAE,MAAM,CAAA;CAAE,CAG9E,CAAA;AAED,eAAO,MAAM,oBAAoB,2BAA2B,CAAA;AAO5D,wEAAwE;AACxE,eAAO,MAAM,OAAO,YAAa,MAAM,KAAG,OAA+C,CAAA;AAEzF,eAAO,MAAM,WAAW,8BAIrB;IACD,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB,IAAI,CAAC,EAAE,MAAM,CAAA;CACd,KAAG,WAmBH,CAAA;AAED,uEAAuE;AACvE,eAAO,MAAM,SAAS,8BAA+B,WAAW,KAAG,MAAwC,CAAA;AAE3G,MAAM,MAAM,KAAK,GAAG,CAAC,GAAG,EAAE,MAAM,EAAE,IAAI,EAAE,WAAW,KAAK,OAAO,CAAC,QAAQ,CAAC,CAAA;AAEzE,0GAA0G;AAC1G,eAAO,MAAM,UAAU,WACb,WAAW,UACX,MAAM,GAAG,SAAS,wCAC2B;IAAE,WAAW,CAAC,EAAE,MAAM,CAAC;IAAC,KAAK,CAAC,EAAE,KAAK,CAAC;IAAC,KAAK,CAAC,EAAE,MAAM,CAAA;CAAE,KAC3G,QAuEF,CAAA"}