@atomic-chat/core 0.19.0 → 0.20.5

package/dist/runtime/geniex/host/engine.d.ts

@@ -0,0 +1,76 @@
+/**
+ * The main thread's handle on the Worker that owns the library (openspec change `add-geniex-runtime`,
+ * design D1): one request at a time, each answered by the Worker's messages, tokens of a generation
+ * delivered as they arrive. The Worker is the web-standard one of the Bun runtime; the tests drive
+ * `WorkerEngine` with an in-process stand-in that runs the Worker's handler on a fake library.
+ */
+import type { GeniexGenerateResult, GeniexGenerationConfig, GeniexModelSettings, GeniexTemplateInput } from './library.js';
+/** What the engine needs of a Worker: the web `Worker` shape, which Bun's is. */
+export interface WorkerLike {
+    postMessage(message: unknown): void;
+    onmessage: ((event: {
+        data: unknown;
+    }) => void) | null;
+    onerror: ((event: {
+        message?: string;
+    }) => void) | null;
+    terminate(): void;
+}
+export interface GeniexProbeAnswer {
+    version: string | null;
+    abi: number;
+    plugins: string[];
+    devices: string[];
+}
+/** The host's asynchronous view of the library; `WorkerEngine` is the one implementation. */
+export interface GeniexEngine {
+    probe(libDir: string): Promise<GeniexProbeAnswer>;
+    create(libDir: string, settings: GeniexModelSettings): Promise<{
+        vision: boolean;
+    }>;
+    applyChatTemplate(input: GeniexTemplateInput): Promise<string>;
+    /**
+     * Tokens arrive through `onToken` as the library makes them; `cancel` (one `Int32`, set non-zero)
+     * ends the generation, which then resolves with `stopReason: 'cancelled'`.
+     */
+    generate(prompt: string, config: GeniexGenerationConfig, onToken: (token: string) => void, cancel: Int32Array): Promise<GeniexGenerateResult>;
+    destroy(): Promise<void>;
+}
+/** The Worker stopped before answering: a crash of the native library takes the Worker with it. */
+export declare class GeniexWorkerGone extends Error {
+    constructor(message: string);
+}
+export declare class WorkerEngine implements GeniexEngine {
+    private readonly worker;
+    private tail;
+    private pending;
+    private onToken;
+    private nextId;
+    private gone;
+    constructor(worker: WorkerLike);
+    private receive;
+    /**
+     * One request, answered by the next non-token message; requests queue behind each other. `before`
+     * runs when the request's turn comes, not when it is queued: a generate's token handler must not
+     * replace the one of a generation still streaming.
+     */
+    private ask;
+    private static refused;
+    probe(libDir: string): Promise<GeniexProbeAnswer>;
+    create(libDir: string, settings: GeniexModelSettings): Promise<{
+        vision: boolean;
+    }>;
+    applyChatTemplate(input: GeniexTemplateInput): Promise<string>;
+    generate(prompt: string, config: GeniexGenerationConfig, onToken: (token: string) => void, cancel: Int32Array): Promise<GeniexGenerateResult>;
+    destroy(): Promise<void>;
+}
+/**
+ * The Worker of the compiled binaries, started from a `blob:` URL over the script that
+ * `scripts/build-geniex-worker.mjs` bundles out of `worker.ts` into `worker-source.generated.ts`.
+ * Not from a path: inside a compiled Windows binary the Bun that CI pins (1.3.10) reads a Worker's
+ * `B:\~BUN\root\worker.js` from the real disk (ENOENT; oven-sh/bun#16869, #29124, fixed in 1.4.1),
+ * and a bare `./worker.js` is resolved against the working directory. The generated module is imported
+ * here and only here, so a checkout without a build still loads this file; under Node there is no
+ * `Worker` global, and no host either.
+ */
+export declare function spawnGeniexWorker(): Promise<WorkerLike>;

package/dist/runtime/geniex/host/engine.js

@@ -0,0 +1,122 @@
+/**
+ * The main thread's handle on the Worker that owns the library (openspec change `add-geniex-runtime`,
+ * design D1): one request at a time, each answered by the Worker's messages, tokens of a generation
+ * delivered as they arrive. The Worker is the web-standard one of the Bun runtime; the tests drive
+ * `WorkerEngine` with an in-process stand-in that runs the Worker's handler on a fake library.
+ */
+import { GeniexLibraryError } from './library.js';
+/** The Worker stopped before answering: a crash of the native library takes the Worker with it. */
+export class GeniexWorkerGone extends Error {
+    constructor(message) {
+        super(message);
+        this.name = 'GeniexWorkerGone';
+    }
+}
+export class WorkerEngine {
+    worker;
+    tail = Promise.resolve();
+    pending = null;
+    onToken = null;
+    nextId = 1;
+    gone = null;
+    constructor(worker) {
+        this.worker = worker;
+        worker.onmessage = (event) => this.receive(event.data);
+        worker.onerror = (event) => {
+            this.gone = new GeniexWorkerGone(`the GenieX worker failed: ${event.message ?? 'unknown error'}`);
+            this.pending?.reject(this.gone);
+            this.pending = null;
+        };
+    }
+    receive(answer) {
+        if (answer.type === 'token') {
+            this.onToken?.(answer.text);
+            return;
+        }
+        const pending = this.pending;
+        this.pending = null;
+        pending?.resolve(answer);
+    }
+    /**
+     * One request, answered by the next non-token message; requests queue behind each other. `before`
+     * runs when the request's turn comes, not when it is queued: a generate's token handler must not
+     * replace the one of a generation still streaming.
+     */
+    ask(request, before) {
+        const run = this.tail.then(() => new Promise((resolve, reject) => {
+            if (this.gone)
+                return reject(this.gone);
+            before?.();
+            this.pending = { resolve, reject };
+            this.worker.postMessage(request);
+        }));
+        this.tail = run.catch(() => { });
+        return run;
+    }
+    static refused(answer, what) {
+        if (answer.type === 'failed' || answer.type === 'error')
+            throw new GeniexLibraryError(answer.type === 'failed' ? answer.stage : what, answer.message, answer.code);
+        throw new GeniexLibraryError(what, `unexpected answer ${answer.type}`);
+    }
+    async probe(libDir) {
+        const answer = await this.ask({ type: 'probe', libDir });
+        if (answer.type !== 'probed')
+            WorkerEngine.refused(answer, 'probe');
+        return { version: answer.version, abi: answer.abi, plugins: answer.plugins, devices: answer.devices };
+    }
+    async create(libDir, settings) {
+        const answer = await this.ask({ type: 'create', libDir, settings });
+        if (answer.type !== 'created')
+            WorkerEngine.refused(answer, 'create');
+        return { vision: answer.vision };
+    }
+    async applyChatTemplate(input) {
+        const answer = await this.ask({ type: 'template', id: this.nextId++, input });
+        if (answer.type !== 'templated')
+            WorkerEngine.refused(answer, 'apply_chat_template');
+        return answer.prompt;
+    }
+    async generate(prompt, config, onToken, cancel) {
+        const buffer = cancel.buffer;
+        if (!(buffer instanceof SharedArrayBuffer))
+            throw new GeniexLibraryError('generate', 'the cancel flag must live in a SharedArrayBuffer');
+        try {
+            const answer = await this.ask({ type: 'generate', id: this.nextId++, prompt, config, cancel: buffer }, () => {
+                this.onToken = onToken;
+            });
+            if (answer.type !== 'done')
+                WorkerEngine.refused(answer, 'generate');
+            return answer.result;
+        }
+        finally {
+            if (this.onToken === onToken)
+                this.onToken = null;
+        }
+    }
+    async destroy() {
+        try {
+            if (!this.gone)
+                await this.ask({ type: 'destroy' });
+        }
+        finally {
+            this.worker.terminate();
+        }
+    }
+}
+/**
+ * The Worker of the compiled binaries, started from a `blob:` URL over the script that
+ * `scripts/build-geniex-worker.mjs` bundles out of `worker.ts` into `worker-source.generated.ts`.
+ * Not from a path: inside a compiled Windows binary the Bun that CI pins (1.3.10) reads a Worker's
+ * `B:\~BUN\root\worker.js` from the real disk (ENOENT; oven-sh/bun#16869, #29124, fixed in 1.4.1),
+ * and a bare `./worker.js` is resolved against the working directory. The generated module is imported
+ * here and only here, so a checkout without a build still loads this file; under Node there is no
+ * `Worker` global, and no host either.
+ */
+export async function spawnGeniexWorker() {
+    const ctor = globalThis.Worker;
+    if (!ctor)
+        throw new GeniexLibraryError('load', 'geniex-host runs only inside a compiled binary (no Worker)');
+    const { GENIEX_WORKER_SOURCE } = await import(/* @vite-ignore */ './worker-source.generated.js');
+    return new ctor(URL.createObjectURL(new Blob([GENIEX_WORKER_SOURCE], { type: 'application/javascript' })));
+}
+//# sourceMappingURL=engine.js.map

package/dist/runtime/geniex/host/engine.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"engine.js","sourceRoot":"","sources":["../../../../src/runtime/geniex/host/engine.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,EAAE,kBAAkB,EAAE,MAAM,cAAc,CAAA;AA0CjD,mGAAmG;AACnG,MAAM,OAAO,gBAAiB,SAAQ,KAAK;IACzC,YAAY,OAAe;QACzB,KAAK,CAAC,OAAO,CAAC,CAAA;QACd,IAAI,CAAC,IAAI,GAAG,kBAAkB,CAAA;IAChC,CAAC;CACF;AAID,MAAM,OAAO,YAAY;IAOM;IANrB,IAAI,GAAqB,OAAO,CAAC,OAAO,EAAE,CAAA;IAC1C,OAAO,GAAmB,IAAI,CAAA;IAC9B,OAAO,GAAqC,IAAI,CAAA;IAChD,MAAM,GAAG,CAAC,CAAA;IACV,IAAI,GAAiB,IAAI,CAAA;IAEjC,YAA6B,MAAkB;QAAlB,WAAM,GAAN,MAAM,CAAY;QAC7C,MAAM,CAAC,SAAS,GAAG,CAAC,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,IAA0B,CAAC,CAAA;QAC5E,MAAM,CAAC,OAAO,GAAG,CAAC,KAAK,EAAE,EAAE;YACzB,IAAI,CAAC,IAAI,GAAG,IAAI,gBAAgB,CAAC,6BAA6B,KAAK,CAAC,OAAO,IAAI,eAAe,EAAE,CAAC,CAAA;YACjG,IAAI,CAAC,OAAO,EAAE,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;YAC/B,IAAI,CAAC,OAAO,GAAG,IAAI,CAAA;QACrB,CAAC,CAAA;IACH,CAAC;IAEO,OAAO,CAAC,MAA0B;QACxC,IAAI,MAAM,CAAC,IAAI,KAAK,OAAO,EAAE,CAAC;YAC5B,IAAI,CAAC,OAAO,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,CAAA;YAC3B,OAAM;QACR,CAAC;QACD,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,CAAA;QAC5B,IAAI,CAAC,OAAO,GAAG,IAAI,CAAA;QACnB,OAAO,EAAE,OAAO,CAAC,MAAM,CAAC,CAAA;IAC1B,CAAC;IAED;;;;OAIG;IACK,GAAG,CAAC,OAA4B,EAAE,MAAmB;QAC3D,MAAM,GAAG,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,CACxB,GAAG,EAAE,CACH,IAAI,OAAO,CAAqB,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;YAClD,IAAI,IAAI,CAAC,IAAI;gBAAE,OAAO,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;YACvC,MAAM,EAAE,EAAE,CAAA;YACV,IAAI,CAAC,OAAO,GAAG,EAAE,OAAO,EAAE,MAAM,EAAE,CAAA;YAClC,IAAI,CAAC,MAAM,CAAC,WAAW,CAAC,OAAO,CAAC,CAAA;QAClC,CAAC,CAAC,CACL,CAAA;QACD,IAAI,CAAC,IAAI,GAAG,GAAG,CAAC,KAAK,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,CAAA;QAC/B,OAAO,GAAG,CAAA;IACZ,CAAC;IAEO,MAAM,CAAC,OAAO,CAAC,MAA0B,EAAE,IAAY;QAC7D,IAAI,MAAM,CAAC,IAAI,KAAK,QAAQ,IAAI,MAAM,CAAC,IAAI,KAAK,OAAO;YACrD,MAAM,IAAI,kBAAkB,CAC1B,MAAM,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,EAC9C,MAAM,CAAC,OAAO,EACd,MAAM,CAAC,IAAI,CACZ,CAAA;QACH,MAAM,IAAI,kBAAkB,CAAC,IAAI,EAAE,qBAAqB,MAAM,CAAC,IAAI,EAAE,CAAC,CAAA;IACxE,CAAC;IAED,KAAK,CAAC,KAAK,CAAC,MAAc;QACxB,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,GAAG,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAA;QACxD,IAAI,MAAM,CAAC,IAAI,KAAK,QAAQ;YAAE,YAAY,CAAC,OAAO,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;QACnE,OAAO,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,EAAE,GAAG,EAAE,MAAM,CAAC,GAAG,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,EAAE,CAAA;IACvG,CAAC;IAED,KAAK,CAAC,MAAM,CAAC,MAAc,EAAE,QAA6B;QACxD,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,GAAG,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,EAAE,QAAQ,EAAE,CAAC,CAAA;QACnE,IAAI,MAAM,CAAC,IAAI,KAAK,SAAS;YAAE,YAAY,CAAC,OAAO,CAAC,MAAM,EAAE,QAAQ,CAAC,CAAA;QACrE,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,CAAA;IAClC,CAAC;IAED,KAAK,CAAC,iBAAiB,CAAC,KAA0B;QAChD,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,GAAG,CAAC,EAAE,IAAI,EAAE,UAAU,EAAE,EAAE,EAAE,IAAI,CAAC,MAAM,EAAE,EAAE,KAAK,EAAE,CAAC,CAAA;QAC7E,IAAI,MAAM,CAAC,IAAI,KAAK,WAAW;YAAE,YAAY,CAAC,OAAO,CAAC,MAAM,EAAE,qBAAqB,CAAC,CAAA;QACpF,OAAO,MAAM,CAAC,MAAM,CAAA;IACtB,CAAC;IAED,KAAK,CAAC,QAAQ,CACZ,MAAc,EACd,MAA8B,EAC9B,OAAgC,EAChC,MAAkB;QAElB,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,CAAA;QAC5B,IAAI,CAAC,CAAC,MAAM,YAAY,iBAAiB,CAAC;YACxC,MAAM,IAAI,kBAAkB,CAAC,UAAU,EAAE,kDAAkD,CAAC,CAAA;QAC9F,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,GAAG,CAC3B,EAAE,IAAI,EAAE,UAAU,EAAE,EAAE,EAAE,IAAI,CAAC,MAAM,EAAE,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,EACvE,GAAG,EAAE;gBACH,IAAI,CAAC,OAAO,GAAG,OAAO,CAAA;YACxB,CAAC,CACF,CAAA;YACD,IAAI,MAAM,CAAC,IAAI,KAAK,MAAM;gBAAE,YAAY,CAAC,OAAO,CAAC,MAAM,EAAE,UAAU,CAAC,CAAA;YACpE,OAAO,MAAM,CAAC,MAAM,CAAA;QACtB,CAAC;gBAAS,CAAC;YACT,IAAI,IAAI,CAAC,OAAO,KAAK,OAAO;gBAAE,IAAI,CAAC,OAAO,GAAG,IAAI,CAAA;QACnD,CAAC;IACH,CAAC;IAED,KAAK,CAAC,OAAO;QACX,IAAI,CAAC;YACH,IAAI,CAAC,IAAI,CAAC,IAAI;gBAAE,MAAM,IAAI,CAAC,GAAG,CAAC,EAAE,IAAI,EAAE,SAAS,EAAE,CAAC,CAAA;QACrD,CAAC;gBAAS,CAAC;YACT,IAAI,CAAC,MAAM,CAAC,SAAS,EAAE,CAAA;QACzB,CAAC;IACH,CAAC;CACF;AAED;;;;;;;;GAQG;AACH,MAAM,CAAC,KAAK,UAAU,iBAAiB;IACrC,MAAM,IAAI,GAAI,UAA2D,CAAC,MAAM,CAAA;IAChF,IAAI,CAAC,IAAI;QACP,MAAM,IAAI,kBAAkB,CAAC,MAAM,EAAE,4DAA4D,CAAC,CAAA;IACpG,MAAM,EAAE,oBAAoB,EAAE,GAAG,MAAM,MAAM,CAAC,kBAAkB,CAAC,8BAA8B,CAAC,CAAA;IAChG,OAAO,IAAI,IAAI,CAAC,GAAG,CAAC,eAAe,CAAC,IAAI,IAAI,CAAC,CAAC,oBAAoB,CAAC,EAAE,EAAE,IAAI,EAAE,wBAAwB,EAAE,CAAC,CAAC,CAAC,CAAA;AAC5G,CAAC"}

package/dist/runtime/geniex/host/ffi-bun.d.ts

@@ -0,0 +1,15 @@
+/**
+ * `GeniexLibrary` over `bun:ffi` — the third file in `src/` that may import `bun:*` (change
+ * `add-geniex-runtime`, design D1; allow-listed in both runtime-agnostic gates with `worker.ts`). It
+ * is imported by the Worker alone, so the npm package still loads under Node.
+ *
+ * The binding follows `sdk/include/geniex.h` of GenieX v0.9.3 (ABI 1 of conf's manifest). Structs
+ * are laid out by hand for a 64-bit little-endian target with natural alignment (Windows arm64 is
+ * the only one); each layout names its offsets next to the C declaration it mirrors. The token
+ * callback is a plain `JSCallback`: the library calls it synchronously on this very thread, from
+ * inside `geniex_llm_generate`, which is why the whole binding lives in the Worker (design D1).
+ *
+ * Unverified against the real library until the spike on a Snapdragon X laptop (rulings/core.md).
+ */
+import type { LoadGeniexLibrary } from './library.js';
+export declare const loadBunGeniexLibrary: LoadGeniexLibrary;

package/dist/runtime/geniex/host/ffi-bun.js

@@ -0,0 +1,415 @@
+/**
+ * `GeniexLibrary` over `bun:ffi` — the third file in `src/` that may import `bun:*` (change
+ * `add-geniex-runtime`, design D1; allow-listed in both runtime-agnostic gates with `worker.ts`). It
+ * is imported by the Worker alone, so the npm package still loads under Node.
+ *
+ * The binding follows `sdk/include/geniex.h` of GenieX v0.9.3 (ABI 1 of conf's manifest). Structs
+ * are laid out by hand for a 64-bit little-endian target with natural alignment (Windows arm64 is
+ * the only one); each layout names its offsets next to the C declaration it mirrors. The token
+ * callback is a plain `JSCallback`: the library calls it synchronously on this very thread, from
+ * inside `geniex_llm_generate`, which is why the whole binding lives in the Worker (design D1).
+ *
+ * Unverified against the real library until the spike on a Snapdragon X laptop (rulings/core.md).
+ */
+import { dlopen, FFIType, JSCallback, ptr, read, CString } from 'bun:ffi';
+import { existsSync } from 'node:fs';
+import { join } from 'node:path';
+import { BOUND_GENIEX_ABI, GENIEX_CANCELLED, GENIEX_POWER_MODES, GeniexLibraryError } from './library.js';
+const T = FFIType;
+const LIBRARY_FILE = 'geniex.dll';
+/** The QAIRT runtime inside a build; told to the library so the `qairt` plugin finds it. */
+const QAIRT_DIR = 'qairt';
+const LOG_LEVEL_ERROR = 4;
+const STOP_REASONS = ['eos', 'length', 'user', 'stop_sequence', 'context_length'];
+const SYMBOLS = {
+    geniex_get_error_message: { args: [T.i32], returns: T.cstring },
+    geniex_init: { args: [], returns: T.i32 },
+    geniex_deinit: { args: [], returns: T.i32 },
+    geniex_set_log_level: { args: [T.i32], returns: T.i32 },
+    geniex_set_qairt_runtime_path: { args: [T.ptr], returns: T.i32 },
+    geniex_free: { args: [T.ptr], returns: T.void },
+    geniex_version: { args: [], returns: T.cstring },
+    geniex_get_plugin_list: { args: [T.ptr], returns: T.i32 },
+    geniex_get_device_list: { args: [T.ptr, T.ptr], returns: T.i32 },
+    geniex_resolve_device: { args: [T.ptr, T.ptr], returns: T.i32 },
+    geniex_llm_create: { args: [T.ptr, T.ptr], returns: T.i32 },
+    geniex_llm_destroy: { args: [T.ptr], returns: T.i32 },
+    geniex_llm_apply_chat_template: { args: [T.ptr, T.ptr, T.ptr], returns: T.i32 },
+    geniex_llm_generate: { args: [T.ptr, T.ptr, T.ptr], returns: T.i32 },
+    geniex_vlm_create: { args: [T.ptr, T.ptr], returns: T.i32 },
+    geniex_vlm_destroy: { args: [T.ptr], returns: T.i32 },
+    geniex_vlm_apply_chat_template: { args: [T.ptr, T.ptr, T.ptr], returns: T.i32 },
+    geniex_vlm_generate: { args: [T.ptr, T.ptr, T.ptr], returns: T.i32 },
+    geniex_vlm_reset: { args: [T.ptr], returns: T.i32 },
+};
+const cstr = (value) => Buffer.from(`${value}\0`, 'utf8');
+const wide = (value) => Buffer.from(`${value}\0`, 'utf16le');
+const text = (value) => {
+    const s = value == null ? '' : String(value);
+    return s === '' ? null : s;
+};
+/** Keeps every buffer a struct points at alive for the length of one library call. */
+class Arena {
+    kept = [];
+    str(value) {
+        if (value === null)
+            return null;
+        const buffer = cstr(value);
+        this.kept.push(buffer);
+        return ptr(buffer);
+    }
+    /** A NUL-terminated array of C strings; `null` for an empty list. */
+    strings(values) {
+        if (values.length === 0)
+            return null;
+        const table = Buffer.alloc(8 * values.length);
+        values.forEach((value, index) => table.writeBigUInt64LE(BigInt(this.str(value) ?? 0), index * 8));
+        this.kept.push(table);
+        return ptr(table);
+    }
+    struct(size) {
+        const buffer = Buffer.alloc(size);
+        this.kept.push(buffer);
+        return buffer;
+    }
+}
+const writePtr = (buffer, offset, value) => buffer.writeBigUInt64LE(BigInt(value ?? 0), offset);
+const readPtr = (buffer, offset) => {
+    const value = buffer.readBigUInt64LE(offset);
+    return value === 0n ? null : Number(value);
+};
+/** Windows searches the library's own folder for its DLLs only when told to. */
+function prepareWindowsSearch(libDir) {
+    const k32 = dlopen('kernel32.dll', { SetDllDirectoryW: { args: [T.ptr], returns: T.i32 } }).symbols;
+    k32.SetDllDirectoryW(ptr(wide(libDir)));
+}
+/**
+ * Windows: out of EcoQoS, as `litert-lm/host/ffi-bun.ts` does. A process with no window counts as
+ * background work and runs on efficiency cores at low clocks; a failure only costs speed.
+ */
+function leaveEcoQos() {
+    const k32 = dlopen('kernel32.dll', {
+        SetProcessInformation: { args: [T.i64, T.i32, T.ptr, T.u32], returns: T.i32 },
+    }).symbols;
+    const PROCESS_POWER_THROTTLING = 4;
+    const state = new Uint32Array([1, 1, 0]);
+    k32.SetProcessInformation(-1n, PROCESS_POWER_THROTTLING, ptr(state), state.byteLength);
+}
+export const loadBunGeniexLibrary = async (libDir) => {
+    const path = join(libDir, LIBRARY_FILE);
+    if (!existsSync(path))
+        throw new GeniexLibraryError('load', `${path} not found: no ${LIBRARY_FILE} in this build`);
+    if (process.platform === 'win32') {
+        prepareWindowsSearch(libDir);
+        leaveEcoQos();
+    }
+    let L;
+    try {
+        L = dlopen(path, SYMBOLS).symbols;
+    }
+    catch (error) {
+        throw new GeniexLibraryError('load', error instanceof Error ? error.message : String(error));
+    }
+    const check = (operation, code) => {
+        const value = Number(code);
+        if (value === 0)
+            return;
+        throw new GeniexLibraryError(operation, text(L.geniex_get_error_message(value)) ?? `error ${value}`, value);
+    };
+    // The QAIRT runtime the plugin needs (QnnHtp.dll, QnnSystem.dll, the HTP stubs and skels) ships in
+    // the SDK under `lib/qairt/htp-files` (spike S4); pinning it there keeps a `GENIEX_QAIRT_LIB` in the
+    // user's environment from swapping it. `lib/qairt` itself holds only the plugin's DLLs.
+    const qairt = join(libDir, QAIRT_DIR, 'htp-files');
+    if (existsSync(join(qairt, 'QnnHtp.dll'))) {
+        const keep = cstr(qairt);
+        check('set_qairt_runtime_path', L.geniex_set_qairt_runtime_path(ptr(keep)));
+    }
+    // Before `geniex_init`, as the header says: the default is TRACE, and the plugin scan at init logs
+    // at that level to stderr (spike S1). The level also sets QNN's in the qairt plugin.
+    L.geniex_set_log_level(LOG_LEVEL_ERROR);
+    check('init', L.geniex_init());
+    return bindLibrary(L, check);
+};
+/** A `const char**` of `count` strings the library allocated; freed after reading. */
+function takeStrings(L, table, count) {
+    if (!table || count <= 0)
+        return [];
+    const out = [];
+    for (let i = 0; i < count; i++) {
+        const at = read.ptr(table, i * 8);
+        out.push(at ? new CString(at).toString() : '');
+    }
+    L.geniex_free(table);
+    return out;
+}
+function bindLibrary(L, check) {
+    // The library says `v0.9.3` (spike S1, 2026-10-28): the `v` is not part of the version.
+    const version = /^\d+\.\d+\.\d+/.exec((text(L.geniex_version()) ?? '').replace(/^v/i, ''))?.[0] ?? null;
+    function plugins() {
+        // geniex_GetPluginListOutput { const char** plugin_ids @0; int32 plugin_count @8 } = 16
+        const out = Buffer.alloc(16);
+        check('get_plugin_list', L.geniex_get_plugin_list(ptr(out)));
+        return takeStrings(L, readPtr(out, 0), out.readInt32LE(8));
+    }
+    function devices(plugin) {
+        const arena = new Arena();
+        // geniex_GetDeviceListInput { const char* plugin_id @0 } = 8
+        const input = arena.struct(8);
+        writePtr(input, 0, arena.str(plugin));
+        // geniex_GetDeviceListOutput { const char** device_ids @0; const char** device_names @8; int32 device_count @16 } = 24
+        const out = arena.struct(24);
+        check('get_device_list', L.geniex_get_device_list(ptr(input), ptr(out)));
+        const count = out.readInt32LE(16);
+        const ids = takeStrings(L, readPtr(out, 0), count);
+        takeStrings(L, readPtr(out, 8), count);
+        return ids;
+    }
+    function resolveDevice(plugin, mode) {
+        const arena = new Arena();
+        // geniex_ResolveDeviceInput { plugin_id @0; model_name @8; mode @16; int32 ngl_default @24 } = 32
+        const input = arena.struct(32);
+        writePtr(input, 0, arena.str(plugin));
+        writePtr(input, 8, null);
+        writePtr(input, 16, arena.str(mode));
+        input.writeInt32LE(999, 24);
+        // geniex_ResolveDeviceOutput { char* device_id @0; int32 ngl @8; char* warning @16 } = 24
+        const out = arena.struct(24);
+        check('resolve_device', L.geniex_resolve_device(ptr(input), ptr(out)));
+        const idPtr = readPtr(out, 0);
+        const warningPtr = readPtr(out, 16);
+        const deviceId = idPtr ? new CString(idPtr).toString() : null;
+        if (idPtr)
+            L.geniex_free(idPtr);
+        if (warningPtr)
+            L.geniex_free(warningPtr);
+        return { deviceId, ngl: out.readInt32LE(8) };
+    }
+    /** geniex_ModelConfig, 80 bytes, written into `buffer` at `at`. */
+    function writeModelConfig(buffer, at, arena, settings, ngl) {
+        // The qairt plugin refuses any n_ctx (S4: `--nctx (n_ctx) is not supported by the qairt plugin`);
+        // the bundle fixes its context, and 0 means "from the model".
+        buffer.writeInt32LE(settings.plugin === 'qairt' ? 0 : settings.nCtx, at + 0); // n_ctx
+        buffer.writeInt32LE(settings.threads, at + 4); // n_threads
+        buffer.writeInt32LE(settings.threads, at + 8); // n_threads_batch
+        buffer.writeInt32LE(0, at + 12); // n_batch
+        buffer.writeInt32LE(0, at + 16); // n_ubatch
+        buffer.writeInt32LE(0, at + 20); // n_seq_max
+        buffer.writeInt32LE(ngl, at + 24); // n_gpu_layers, as resolve_device said
+        writePtr(buffer, at + 32, null); // chat_template_path
+        writePtr(buffer, at + 40, null); // chat_template_content
+        writePtr(buffer, at + 48, arena.str('')); // spec_type: disabled
+        writePtr(buffer, at + 56, arena.str('')); // spec_draft_model
+        buffer.writeInt32LE(0, at + 64); // spec_n_max
+        buffer.writeInt32LE(0, at + 68); // spec_n_min
+        buffer.writeFloatLE(0, at + 72); // spec_p_min
+        buffer.writeInt32LE(GENIEX_POWER_MODES.indexOf(settings.powerMode), at + 76); // power_mode
+    }
+    function createModel(settings) {
+        const { deviceId, ngl } = resolveDevice(settings.plugin, settings.device);
+        const vision = settings.mmproj !== null;
+        // The qairt plugin resolves a bundle from the folder of the FILE it is given (spike S4: handed the
+        // folder, it looked for tokenizer.json in the parent); `genie_config.json` is the file every bundle has.
+        const modelPath = settings.plugin === 'qairt' ? join(settings.model, 'genie_config.json') : settings.model;
+        const arena = new Arena();
+        const handleOut = arena.struct(8);
+        if (vision) {
+            // geniex_VlmCreateInput { model_path @0; mmproj_path @8; config @16 (80); plugin_id @96; device_id @104; vit_device_id @112; tokenizer_path @120 } = 128
+            const input = arena.struct(128);
+            writePtr(input, 0, arena.str(modelPath));
+            writePtr(input, 8, arena.str(settings.mmproj));
+            writeModelConfig(input, 16, arena, settings, ngl);
+            writePtr(input, 96, arena.str(settings.plugin));
+            writePtr(input, 104, arena.str(deviceId));
+            writePtr(input, 112, null);
+            writePtr(input, 120, null);
+            check('vlm_create', L.geniex_vlm_create(ptr(input), ptr(handleOut)));
+        }
+        else {
+            // geniex_LlmCreateInput { model_path @0; tokenizer_path @8; config @16 (80); plugin_id @96; device_id @104 } = 112
+            const input = arena.struct(112);
+            writePtr(input, 0, arena.str(modelPath));
+            writePtr(input, 8, null);
+            writeModelConfig(input, 16, arena, settings, ngl);
+            writePtr(input, 96, arena.str(settings.plugin));
+            writePtr(input, 104, arena.str(deviceId));
+            check('llm_create', L.geniex_llm_create(ptr(input), ptr(handleOut)));
+        }
+        const handle = readPtr(handleOut, 0);
+        if (!handle)
+            throw new GeniexLibraryError(vision ? 'vlm_create' : 'llm_create', 'the library returned no handle');
+        return bindModel(L, check, handle, vision);
+    }
+    return {
+        version,
+        abi: BOUND_GENIEX_ABI,
+        plugins,
+        devices,
+        createModel,
+        close: () => {
+            L.geniex_deinit();
+        },
+    };
+}
+/** geniex_ToolCall[] { id @0; name @8; arguments @16 } = 24 each. */
+function writeToolCalls(arena, calls) {
+    if (calls.length === 0)
+        return null;
+    const table = arena.struct(24 * calls.length);
+    calls.forEach((call, index) => {
+        writePtr(table, index * 24, arena.str(call.id));
+        writePtr(table, index * 24 + 8, arena.str(call.name));
+        writePtr(table, index * 24 + 16, arena.str(call.arguments));
+    });
+    return ptr(table);
+}
+function bindModel(L, check, handle, vision) {
+    function applyChatTemplate(input) {
+        const arena = new Arena();
+        const count = input.messages.length;
+        let messages;
+        if (vision) {
+            // geniex_VlmChatMessage { role @0; contents @8; int64 content_count @16; tool_calls @24; int32 tool_call_count @32; tool_call_id @40; tool_name @48 } = 56
+            messages = arena.struct(56 * count);
+            input.messages.forEach((message, index) => {
+                const at = index * 56;
+                // geniex_VlmContent[] { type @0; text @8 } = 16 each
+                const contents = arena.struct(16 * Math.max(1, message.content.length));
+                message.content.forEach((part, j) => {
+                    writePtr(contents, j * 16, arena.str(part.type));
+                    writePtr(contents, j * 16 + 8, arena.str(part.type === 'text' ? part.text : part.path));
+                });
+                writePtr(messages, at, arena.str(message.role));
+                writePtr(messages, at + 8, message.content.length > 0 ? ptr(contents) : null);
+                messages.writeBigInt64LE(BigInt(message.content.length), at + 16);
+                writePtr(messages, at + 24, writeToolCalls(arena, message.tool_calls ?? []));
+                messages.writeInt32LE(message.tool_calls?.length ?? 0, at + 32);
+                writePtr(messages, at + 40, arena.str(message.tool_call_id ?? null));
+                writePtr(messages, at + 48, arena.str(message.tool_name ?? null));
+            });
+        }
+        else {
+            // geniex_LlmChatMessage { role @0; content @8; tool_calls @16; int32 tool_call_count @24; tool_call_id @32; tool_name @40 } = 48
+            messages = arena.struct(48 * count);
+            input.messages.forEach((message, index) => {
+                const at = index * 48;
+                const content = message.content.map((part) => (part.type === 'text' ? part.text : '')).join('');
+                writePtr(messages, at, arena.str(message.role));
+                writePtr(messages, at + 8, arena.str(content));
+                writePtr(messages, at + 16, writeToolCalls(arena, message.tool_calls ?? []));
+                messages.writeInt32LE(message.tool_calls?.length ?? 0, at + 24);
+                writePtr(messages, at + 32, arena.str(message.tool_call_id ?? null));
+                writePtr(messages, at + 40, arena.str(message.tool_name ?? null));
+            });
+        }
+        // Llm: { messages @0; int32 message_count @8; tools @16; bool enable_thinking @24; bool add_generation_prompt @25 } = 32
+        // Vlm: { messages @0; int32 message_count @8; tools @16; bool enable_thinking @24; bool grounding @25 } = 32
+        const request = arena.struct(32);
+        writePtr(request, 0, count > 0 ? ptr(messages) : null);
+        request.writeInt32LE(count, 8);
+        writePtr(request, 16, arena.str(input.tools));
+        request.writeUInt8(input.enableThinking ? 1 : 0, 24);
+        request.writeUInt8(vision ? 0 : 1, 25);
+        // Output { char* formatted_text @0 } = 8
+        const out = arena.struct(8);
+        check(vision ? 'vlm_apply_chat_template' : 'llm_apply_chat_template', vision
+            ? L.geniex_vlm_apply_chat_template(handle, ptr(request), ptr(out))
+            : L.geniex_llm_apply_chat_template(handle, ptr(request), ptr(out)));
+        const formatted = readPtr(out, 0);
+        if (!formatted)
+            return '';
+        const prompt = new CString(formatted).toString();
+        L.geniex_free(formatted);
+        return prompt;
+    }
+    function generate(prompt, config, onToken) {
+        const arena = new Arena();
+        // geniex_SamplerConfig { f32 temperature @0; f32 top_p @4; i32 top_k @8; f32 min_p @12; f32 repetition_penalty @16;
+        //   f32 presence_penalty @20; f32 frequency_penalty @24; i32 seed @28; grammar_path @32; grammar_string @40; bool ignore_eos @48 } = 56
+        const sampler = arena.struct(56);
+        const s = config.sampler;
+        sampler.writeFloatLE(s.temperature ?? 0.8, 0);
+        sampler.writeFloatLE(s.topP ?? 0.95, 4);
+        sampler.writeInt32LE(s.topK ?? 40, 8);
+        sampler.writeFloatLE(0.05, 12);
+        sampler.writeFloatLE(1.0, 16);
+        sampler.writeFloatLE(0, 20);
+        sampler.writeFloatLE(0, 24);
+        sampler.writeInt32LE(s.seed ?? -1, 28);
+        writePtr(sampler, 32, null);
+        writePtr(sampler, 40, arena.str(s.grammar));
+        sampler.writeUInt8(0, 48);
+        // geniex_GenerationConfig { i32 max_tokens @0; stop @8; i32 stop_count @16; sampler_config @24; image_paths @32;
+        //   i32 image_count @40; audio_paths @48; i32 audio_count @56; bool sliding_window @60; i32 sliding_window_n_keep @64 } = 72
+        const generation = arena.struct(72);
+        generation.writeInt32LE(config.maxTokens, 0);
+        writePtr(generation, 8, arena.strings(config.stop));
+        generation.writeInt32LE(config.stop.length, 16);
+        writePtr(generation, 24, ptr(sampler));
+        writePtr(generation, 32, arena.strings(config.imagePaths));
+        generation.writeInt32LE(config.imagePaths.length, 40);
+        writePtr(generation, 48, null);
+        generation.writeInt32LE(0, 56);
+        generation.writeUInt8(0, 60);
+        generation.writeInt32LE(0, 64);
+        let callbackError = null;
+        const callback = new JSCallback((tokenPtr) => {
+            try {
+                return onToken(tokenPtr ? new CString(tokenPtr).toString() : '');
+            }
+            catch (error) {
+                callbackError = error;
+                return false;
+            }
+        }, { args: [T.ptr, T.ptr], returns: T.bool });
+        try {
+            // Llm: { prompt_utf8 @0; config @8; on_token @16; user_data @24; input_ids @32; i32 input_ids_count @40; input_embd @48; i32 count @56; i32 dim @60 } = 64
+            // Vlm: { prompt_utf8 @0; config @8; on_token @16; user_data @24 } = 32 (the Llm layout is a superset, zero beyond)
+            const input = arena.struct(64);
+            writePtr(input, 0, arena.str(prompt));
+            writePtr(input, 8, ptr(generation));
+            writePtr(input, 16, callback.ptr);
+            writePtr(input, 24, null);
+            // Output { char* full_text @0; geniex_ProfileData @8: i64 ttft @8, media_time @16, prompt_time @24, decode_time @32,
+            //   prompt_tokens @40, generated_tokens @48, f64 prefill_speed @56, decoding_speed @64, i64 draft_n_total @72, draft_n_accepted @80, stop_reason @88 } = 96
+            const out = arena.struct(96);
+            // The llama_cpp VLM keeps the last generation's prefix and refuses a prompt that diverges from it
+            // (S2 on Gemma 4: `VLM prefix reuse failed: prompt[12:18] does not match last generation`);
+            // every request here is a full prompt, so the state is cleared first.
+            if (vision)
+                check('vlm_reset', L.geniex_vlm_reset(handle));
+            const code = Number(vision
+                ? L.geniex_vlm_generate(handle, ptr(input), ptr(out))
+                : L.geniex_llm_generate(handle, ptr(input), ptr(out)));
+            if (callbackError)
+                throw callbackError;
+            if (code === GENIEX_CANCELLED)
+                throw new GeniexLibraryError(vision ? 'vlm_generate' : 'llm_generate', 'Operation cancelled by caller', code);
+            check(vision ? 'vlm_generate' : 'llm_generate', code);
+            const fullText = readPtr(out, 0);
+            if (fullText)
+                L.geniex_free(fullText);
+            const stopPtr = readPtr(out, 88);
+            const stop = stopPtr ? new CString(stopPtr).toString() : '';
+            return {
+                stopReason: STOP_REASONS.find((reason) => reason === stop) ?? 'unknown',
+                promptTokens: Number(out.readBigInt64LE(40)),
+                generatedTokens: Number(out.readBigInt64LE(48)),
+            };
+        }
+        finally {
+            callback.close();
+        }
+    }
+    return {
+        vision,
+        applyChatTemplate,
+        generate,
+        destroy: () => {
+            if (vision)
+                L.geniex_vlm_destroy(handle);
+            else
+                L.geniex_llm_destroy(handle);
+        },
+    };
+}
+//# sourceMappingURL=ffi-bun.js.map