@elizaos/plugin-local-inference 2.0.0-beta.1 → 2.0.3-beta.3

package/dist/adapters/capacitor-llama/types.d.ts

@@ -0,0 +1,338 @@
+/**
+ * Canonical Capacitor-llama adapter contract.
+ *
+ * This is the SAME shape used by `llama-cpp-capacitor` on mobile (iOS / Android,
+ * including riscv64 once Wave 2's libllama.so is installed) and by the bun:ffi
+ * backend on desktop. Every entry point in this plugin consumes one of these
+ * interfaces — there is no `node-llama-cpp` import anywhere downstream.
+ *
+ * Shape rationale: the interface mirrors `llama-cpp-capacitor`'s
+ * `LlamaContext` class verbatim so the mobile path can implement it with a
+ * straight pass-through. The loader is mobile-only — libllama is retired, so
+ * there is no desktop façade (desktop runs the fused `libelizainference`).
+ *
+ * Why a TS interface rather than `import type { LlamaContext } from
+ * "llama-cpp-capacitor"`? The Capacitor package is mobile-only at install time
+ * (Capacitor SPI'd Java/Swift bindings). Importing its types from the desktop
+ * path would force every desktop tsc run to pull the mobile binding's d.ts. We
+ * keep the interface free-standing so any backend can implement it.
+ */
+export interface TokenizerConfig {
+    name: string;
+    type: string;
+}
+export interface CapacitorLlamaContextParams {
+    /** GGUF model path (absolute or Capacitor-asset URI). */
+    model: string;
+    /** Optional Jinja chat template override. */
+    chat_template?: string;
+    /** When loading from a packaged Android/iOS asset. */
+    is_model_asset?: boolean;
+    use_progress_callback?: boolean;
+    n_ctx?: number;
+    n_batch?: number;
+    n_ubatch?: number;
+    n_threads?: number;
+    /**
+     * GPU layer offload. `999` = "all layers on GPU" per llama.cpp convention.
+     * The Capacitor binding rejects > model layer count; desktop path clamps
+     * to the model's metadata layer count.
+     */
+    n_gpu_layers?: number;
+    no_gpu_devices?: boolean;
+    flash_attn?: boolean;
+    cache_type_k?: "f16" | "f32" | "q8_0" | "q4_0" | "q4_1" | "iq4_nl" | "q5_0" | "q5_1";
+    cache_type_v?: "f16" | "f32" | "q8_0" | "q4_0" | "q4_1" | "iq4_nl" | "q5_0" | "q5_1";
+    use_mlock?: boolean;
+    use_mmap?: boolean;
+    vocab_only?: boolean;
+    lora?: string;
+    lora_scaled?: number;
+    lora_list?: Array<{
+        path: string;
+        scaled?: number;
+    }>;
+    rope_freq_base?: number;
+    rope_freq_scale?: number;
+    pooling_type?: "none" | "mean" | "cls" | "last" | "rank";
+    ctx_shift?: boolean;
+    kv_unified?: boolean;
+    swa_full?: boolean;
+    n_cpu_moe?: number;
+    embedding?: boolean;
+    embd_normalize?: number;
+}
+export interface CapacitorLlamaMessagePart {
+    type: string;
+    text?: string;
+    image_url?: {
+        url?: string;
+    };
+    input_audio?: {
+        format: string;
+        data?: string;
+        url?: string;
+    };
+}
+export interface CapacitorLlamaChatMessage {
+    role: string;
+    content?: string | CapacitorLlamaMessagePart[];
+}
+export interface CapacitorLlamaResponseFormat {
+    type: "text" | "json_object" | "json_schema";
+    json_schema?: {
+        strict?: boolean;
+        schema: object;
+    };
+    schema?: object;
+}
+export interface CapacitorLlamaCompletionParams {
+    /** Raw prompt string. Mutually exclusive with `messages`. */
+    prompt?: string;
+    /** OpenAI-style messages; the binding renders the chat template. */
+    messages?: CapacitorLlamaChatMessage[];
+    chat_template?: string;
+    jinja?: boolean;
+    json_schema?: string;
+    /** GBNF grammar source. */
+    grammar?: string;
+    grammar_lazy?: boolean;
+    grammar_triggers?: Array<{
+        type: number;
+        value: string;
+        token: number;
+    }>;
+    enable_thinking?: boolean;
+    thinking_forced_open?: boolean;
+    preserved_tokens?: string[];
+    chat_format?: number;
+    reasoning_format?: string;
+    media_paths?: string[];
+    stop?: string[];
+    n_predict?: number;
+    n_probs?: number;
+    top_k?: number;
+    top_p?: number;
+    min_p?: number;
+    xtc_probability?: number;
+    xtc_threshold?: number;
+    typical_p?: number;
+    temperature?: number;
+    penalty_last_n?: number;
+    penalty_repeat?: number;
+    penalty_freq?: number;
+    penalty_present?: number;
+    mirostat?: number;
+    mirostat_tau?: number;
+    mirostat_eta?: number;
+    dry_multiplier?: number;
+    dry_base?: number;
+    dry_allowed_length?: number;
+    dry_penalty_last_n?: number;
+    dry_sequence_breakers?: string[];
+    top_n_sigma?: number;
+    ignore_eos?: boolean;
+    logit_bias?: Array<[number, number]>;
+    seed?: number;
+    guide_tokens?: number[];
+    tools?: object;
+    parallel_tool_calls?: object;
+    tool_choice?: string;
+    response_format?: CapacitorLlamaResponseFormat;
+    add_generation_prompt?: boolean;
+    chat_template_kwargs?: Record<string, string>;
+    prefill_text?: string;
+}
+export interface CapacitorLlamaToolCall {
+    type: "function";
+    id?: string;
+    function: {
+        name: string;
+        arguments: string;
+    };
+}
+export interface CapacitorLlamaTokenData {
+    token: string;
+    completion_probabilities?: Array<{
+        content: string;
+        probs: Array<{
+            tok_str: string;
+            prob: number;
+        }>;
+    }>;
+    content?: string;
+    reasoning_content?: string;
+    tool_calls?: CapacitorLlamaToolCall[];
+    accumulated_text?: string;
+}
+export interface CapacitorLlamaCompletionResult {
+    text: string;
+    reasoning_content: string;
+    tool_calls: CapacitorLlamaToolCall[];
+    content: string;
+    chat_format: number;
+    tokens_predicted: number;
+    tokens_evaluated: number;
+    truncated: boolean;
+    stopped_eos: boolean;
+    stopped_word: string;
+    stopped_limit: number;
+    stopping_word: string;
+    context_full: boolean;
+    interrupted: boolean;
+    tokens_cached: number;
+    timings: {
+        prompt_n: number;
+        prompt_ms: number;
+        prompt_per_token_ms: number;
+        prompt_per_second: number;
+        predicted_n: number;
+        predicted_ms: number;
+        predicted_per_token_ms: number;
+        predicted_per_second: number;
+    };
+    completion_probabilities?: Array<{
+        content: string;
+        probs: Array<{
+            tok_str: string;
+            prob: number;
+        }>;
+    }>;
+    audio_tokens?: number[];
+}
+export interface CapacitorLlamaTokenizeResult {
+    tokens: number[];
+    has_images: boolean;
+    bitmap_hashes: number[];
+    chunk_pos: number[];
+    chunk_pos_images: number[];
+}
+export interface CapacitorLlamaEmbeddingResult {
+    embedding: number[];
+}
+export interface CapacitorLlamaBenchResult {
+    modelDesc: string;
+    modelSize: number;
+    modelNParams: number;
+    ppAvg: number;
+    ppStd: number;
+    tgAvg: number;
+    tgStd: number;
+}
+export interface CapacitorLlamaModelDescriptor {
+    desc: string;
+    size: number;
+    nEmbd: number;
+    nParams: number;
+    chatTemplates: {
+        llamaChat: boolean;
+        minja: {
+            default: boolean;
+            defaultCaps: {
+                tools: boolean;
+                toolCalls: boolean;
+                toolResponses: boolean;
+                systemRole: boolean;
+                parallelToolCalls: boolean;
+                toolCallId: boolean;
+            };
+            toolUse: boolean;
+            toolUseCaps: {
+                tools: boolean;
+                toolCalls: boolean;
+                toolResponses: boolean;
+                systemRole: boolean;
+                parallelToolCalls: boolean;
+                toolCallId: boolean;
+            };
+        };
+    };
+    metadata: object;
+    isChatTemplateSupported: boolean;
+}
+/**
+ * The canonical context handle every backend implements. Designed so a single
+ * caller can switch between `llama-cpp-capacitor`'s `LlamaContext` and the
+ * desktop bun:ffi-backed implementation with zero behavioural change at the
+ * TypeScript level.
+ *
+ * Backends that genuinely cannot implement a method (e.g. `bench()` on the
+ * desktop FFI adapter v1) MUST throw a `CapacitorLlamaUnsupportedError` so
+ * callers can route around the gap explicitly rather than silently producing
+ * wrong results.
+ */
+export interface CapacitorLlamaContext {
+    readonly id: number;
+    readonly gpu: boolean;
+    readonly reasonNoGPU: string;
+    readonly model: CapacitorLlamaModelDescriptor;
+    completion(params: CapacitorLlamaCompletionParams, callback?: (data: CapacitorLlamaTokenData) => void): Promise<CapacitorLlamaCompletionResult>;
+    stopCompletion(): Promise<void>;
+    tokenize(text: string, options?: {
+        media_paths?: string[];
+    }): Promise<CapacitorLlamaTokenizeResult>;
+    detokenize(tokens: number[]): Promise<string>;
+    embedding(text: string, params?: {
+        embd_normalize?: number;
+    }): Promise<CapacitorLlamaEmbeddingResult>;
+    bench(pp: number, tg: number, pl: number, nr: number): Promise<CapacitorLlamaBenchResult>;
+    release(): Promise<void>;
+}
+/** Thrown by adapter methods that aren't implemented for a given backend. */
+export declare class CapacitorLlamaUnsupportedError extends Error {
+    readonly method: string;
+    readonly backend: "mobile" | "desktop-ffi";
+    constructor(method: string, backend: "mobile" | "desktop-ffi", message?: string);
+}
+export interface ModelSpec {
+    name: string;
+    repo: string;
+    size: string;
+    quantization: string;
+    contextSize: number;
+    tokenizer: TokenizerConfig;
+}
+export interface EmbeddingModelSpec extends ModelSpec {
+    dimensions: number;
+}
+/**
+ * Default model bundle. Vision and TTS are owned by `plugin-local-inference`'s
+ * voice and vision subsystems — they don't live on this adapter.
+ */
+export declare const MODEL_SPECS: {
+    readonly small: {
+        readonly name: "text/eliza-1-2b-128k.gguf";
+        readonly repo: "elizaos/eliza-1";
+        readonly size: "2B";
+        readonly quantization: "fused GGUF";
+        readonly contextSize: 131072;
+        readonly tokenizer: {
+            readonly name: "elizaos/eliza-1";
+            readonly type: "eliza1";
+        };
+    };
+    readonly medium: {
+        readonly name: "text/eliza-1-4b-128k.gguf";
+        readonly repo: "elizaos/eliza-1";
+        readonly size: "4B";
+        readonly quantization: "fused GGUF";
+        readonly contextSize: 131072;
+        readonly tokenizer: {
+            readonly name: "elizaos/eliza-1";
+            readonly type: "eliza1";
+        };
+    };
+    readonly embedding: {
+        readonly name: "gte-small_fp16.gguf";
+        readonly repo: "ChristianAzinn/gte-small-gguf";
+        readonly size: "64 MB";
+        readonly quantization: "fp16 GGUF";
+        readonly contextSize: 512;
+        readonly dimensions: 384;
+        readonly tokenizer: {
+            readonly name: "ChristianAzinn/gte-small-gguf";
+            readonly type: "bert";
+        };
+    };
+};
+//# sourceMappingURL=types.d.ts.map

package/dist/adapters/capacitor-llama/types.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../../src/adapters/capacitor-llama/types.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAEH,MAAM,WAAW,eAAe;IAC/B,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;CACb;AAID,MAAM,WAAW,2BAA2B;IAC3C,yDAAyD;IACzD,KAAK,EAAE,MAAM,CAAC;IACd,6CAA6C;IAC7C,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,sDAAsD;IACtD,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,qBAAqB,CAAC,EAAE,OAAO,CAAC;IAChC,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB;;;;OAIG;IACH,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,YAAY,CAAC,EACV,KAAK,GACL,KAAK,GACL,MAAM,GACN,MAAM,GACN,MAAM,GACN,QAAQ,GACR,MAAM,GACN,MAAM,CAAC;IACV,YAAY,CAAC,EACV,KAAK,GACL,KAAK,GACL,MAAM,GACN,MAAM,GACN,MAAM,GACN,QAAQ,GACR,MAAM,GACN,MAAM,CAAC;IACV,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,MAAM,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IACrD,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,YAAY,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG,KAAK,GAAG,MAAM,GAAG,MAAM,CAAC;IACzD,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,cAAc,CAAC,EAAE,MAAM,CAAC;CACxB;AAED,MAAM,WAAW,yBAAyB;IACzC,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,SAAS,CAAC,EAAE;QAAE,GAAG,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;IAC7B,WAAW,CAAC,EAAE;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,IAAI,CAAC,EAAE,MAAM,CAAC;QAAC,GAAG,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;CAC9D;AAED,MAAM,WAAW,yBAAyB;IACzC,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,CAAC,EAAE,MAAM,GAAG,yBAAyB,EAAE,CAAC;CAC/C;AAED,MAAM,WAAW,4BAA4B;IAC5C,IAAI,EAAE,MAAM,GAAG,aAAa,GAAG,aAAa,CAAC;IAC7C,WAAW,CAAC,EAAE;QAAE,MAAM,CAAC,EAAE,OAAO,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,CAAC;IACnD,MAAM,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,8BAA8B;IAC9C,6DAA6D;IAC7D,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,oEAAoE;IACpE,QAAQ,CAAC,EAAE,yBAAyB,EAAE,CAAC;IACvC,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,2BAA2B;IAC3B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,gBAAgB,CAAC,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IACzE,eAAe,CAAC,EAAE,OAAO,CAAC;IAC1B,oBAAoB,CAAC,EAAE,OAAO,CAAC;IAC/B,gBAAgB,CAAC,EAAE,MAAM,EAAE,CAAC;IAC5B,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,WAAW,CAAC,EAAE,MAAM,EAAE,CAAC;IACvB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,qBAAqB,CAAC,EAAE,MAAM,EAAE,CAAC;IACjC,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,UAAU,CAAC,EAAE,KAAK,CAAC,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,CAAC;IACrC,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;IACxB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,eAAe,CAAC,EAAE,4BAA4B,CAAC;IAC/C,qBAAqB,CAAC,EAAE,OAAO,CAAC;IAChC,oBAAoB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC9C,YAAY,CAAC,EAAE,MAAM,CAAC;CACtB;AAED,MAAM,WAAW,sBAAsB;IACtC,IAAI,EAAE,UAAU,CAAC;IACjB,EAAE,CAAC,EAAE,MAAM,CAAC;IACZ,QAAQ,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,MAAM,CAAA;KAAE,CAAC;CAC9C;AAED,MAAM,WAAW,uBAAuB;IACvC,KAAK,EAAE,MAAM,CAAC;IACd,wBAAwB,CAAC,EAAE,KAAK,CAAC;QAChC,OAAO,EAAE,MAAM,CAAC;QAChB,KAAK,EAAE,KAAK,CAAC;YAAE,OAAO,EAAE,MAAM,CAAC;YAAC,IAAI,EAAE,MAAM,CAAA;SAAE,CAAC,CAAC;KAChD,CAAC,CAAC;IACH,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,UAAU,CAAC,EAAE,sBAAsB,EAAE,CAAC;IACtC,gBAAgB,CAAC,EAAE,MAAM,CAAC;CAC1B;AAED,MAAM,WAAW,8BAA8B;IAC9C,IAAI,EAAE,MAAM,CAAC;IACb,iBAAiB,EAAE,MAAM,CAAC;IAC1B,UAAU,EAAE,sBAAsB,EAAE,CAAC;IACrC,OAAO,EAAE,MAAM,CAAC;IAChB,WAAW,EAAE,MAAM,CAAC;IACpB,gBAAgB,EAAE,MAAM,CAAC;IACzB,gBAAgB,EAAE,MAAM,CAAC;IACzB,SAAS,EAAE,OAAO,CAAC;IACnB,WAAW,EAAE,OAAO,CAAC;IACrB,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,aAAa,EAAE,MAAM,CAAC;IACtB,YAAY,EAAE,OAAO,CAAC;IACtB,WAAW,EAAE,OAAO,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,OAAO,EAAE;QACR,QAAQ,EAAE,MAAM,CAAC;QACjB,SAAS,EAAE,MAAM,CAAC;QAClB,mBAAmB,EAAE,MAAM,CAAC;QAC5B,iBAAiB,EAAE,MAAM,CAAC;QAC1B,WAAW,EAAE,MAAM,CAAC;QACpB,YAAY,EAAE,MAAM,CAAC;QACrB,sBAAsB,EAAE,MAAM,CAAC;QAC/B,oBAAoB,EAAE,MAAM,CAAC;KAC7B,CAAC;IACF,wBAAwB,CAAC,EAAE,KAAK,CAAC;QAChC,OAAO,EAAE,MAAM,CAAC;QAChB,KAAK,EAAE,KAAK,CAAC;YAAE,OAAO,EAAE,MAAM,CAAC;YAAC,IAAI,EAAE,MAAM,CAAA;SAAE,CAAC,CAAC;KAChD,CAAC,CAAC;IACH,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;CACxB;AAED,MAAM,WAAW,4BAA4B;IAC5C,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,UAAU,EAAE,OAAO,CAAC;IACpB,aAAa,EAAE,MAAM,EAAE,CAAC;IACxB,SAAS,EAAE,MAAM,EAAE,CAAC;IACpB,gBAAgB,EAAE,MAAM,EAAE,CAAC;CAC3B;AAED,MAAM,WAAW,6BAA6B;IAC7C,SAAS,EAAE,MAAM,EAAE,CAAC;CACpB;AAED,MAAM,WAAW,yBAAyB;IACzC,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,YAAY,EAAE,MAAM,CAAC;IACrB,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;CACd;AAID,MAAM,WAAW,6BAA6B;IAC7C,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,MAAM,CAAC;IAChB,aAAa,EAAE;QACd,SAAS,EAAE,OAAO,CAAC;QACnB,KAAK,EAAE;YACN,OAAO,EAAE,OAAO,CAAC;YACjB,WAAW,EAAE;gBACZ,KAAK,EAAE,OAAO,CAAC;gBACf,SAAS,EAAE,OAAO,CAAC;gBACnB,aAAa,EAAE,OAAO,CAAC;gBACvB,UAAU,EAAE,OAAO,CAAC;gBACpB,iBAAiB,EAAE,OAAO,CAAC;gBAC3B,UAAU,EAAE,OAAO,CAAC;aACpB,CAAC;YACF,OAAO,EAAE,OAAO,CAAC;YACjB,WAAW,EAAE;gBACZ,KAAK,EAAE,OAAO,CAAC;gBACf,SAAS,EAAE,OAAO,CAAC;gBACnB,aAAa,EAAE,OAAO,CAAC;gBACvB,UAAU,EAAE,OAAO,CAAC;gBACpB,iBAAiB,EAAE,OAAO,CAAC;gBAC3B,UAAU,EAAE,OAAO,CAAC;aACpB,CAAC;SACF,CAAC;KACF,CAAC;IACF,QAAQ,EAAE,MAAM,CAAC;IACjB,uBAAuB,EAAE,OAAO,CAAC;CACjC;AAID;;;;;;;;;;GAUG;AACH,MAAM,WAAW,qBAAqB;IACrC,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC;IACpB,QAAQ,CAAC,GAAG,EAAE,OAAO,CAAC;IACtB,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,KAAK,EAAE,6BAA6B,CAAC;IAE9C,UAAU,CACT,MAAM,EAAE,8BAA8B,EACtC,QAAQ,CAAC,EAAE,CAAC,IAAI,EAAE,uBAAuB,KAAK,IAAI,GAChD,OAAO,CAAC,8BAA8B,CAAC,CAAC;IAE3C,cAAc,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC;IAEhC,QAAQ,CACP,IAAI,EAAE,MAAM,EACZ,OAAO,CAAC,EAAE;QAAE,WAAW,CAAC,EAAE,MAAM,EAAE,CAAA;KAAE,GAClC,OAAO,CAAC,4BAA4B,CAAC,CAAC;IAEzC,UAAU,CAAC,MAAM,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC;IAE9C,SAAS,CACR,IAAI,EAAE,MAAM,EACZ,MAAM,CAAC,EAAE;QAAE,cAAc,CAAC,EAAE,MAAM,CAAA;KAAE,GAClC,OAAO,CAAC,6BAA6B,CAAC,CAAC;IAE1C,KAAK,CACJ,EAAE,EAAE,MAAM,EACV,EAAE,EAAE,MAAM,EACV,EAAE,EAAE,MAAM,EACV,EAAE,EAAE,MAAM,GACR,OAAO,CAAC,yBAAyB,CAAC,CAAC;IAEtC,OAAO,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC;CACzB;AAED,6EAA6E;AAC7E,qBAAa,8BAA+B,SAAQ,KAAK;IAEvD,QAAQ,CAAC,MAAM,EAAE,MAAM;IACvB,QAAQ,CAAC,OAAO,EAAE,QAAQ,GAAG,aAAa;gBADjC,MAAM,EAAE,MAAM,EACd,OAAO,EAAE,QAAQ,GAAG,aAAa,EAC1C,OAAO,CAAC,EAAE,MAAM;CAQjB;AAID,MAAM,WAAW,SAAS;IACzB,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,YAAY,EAAE,MAAM,CAAC;IACrB,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,EAAE,eAAe,CAAC;CAC3B;AAED,MAAM,WAAW,kBAAmB,SAAQ,SAAS;IACpD,UAAU,EAAE,MAAM,CAAC;CACnB;AAED;;;GAGG;AACH,eAAO,MAAM,WAAW;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CA8BvB,CAAC"}

package/dist/adapters/capacitor-llama/voice-turn.d.ts

@@ -0,0 +1,86 @@
+/**
+ * On-device fused voice-turn entry for the Capacitor / AOSP device bridge.
+ *
+ * This is the production caller for `LocalInferenceEngine.runVoiceTurn` (#8786):
+ * the in-process fused mic→speech loop (ASR → {MTP drafts ∥ target verify} →
+ * phrase chunker → OmniVoice/Kokoro → PCM ring buffer) that the engine's voice
+ * bridge owns. The engine bridge serves ASR + TTS from the fused
+ * `libelizainference` handle; the DEVICE supplies its own text runner so the
+ * draft/verify loop runs on the on-device text model (the
+ * `CapacitorLlamaContext`-backed decode) rather than the desktop dispatcher —
+ * exactly the seam `runVoiceTurn`'s `textRunner` option was built for.
+ *
+ * The native iOS/Android mic-capture layer hands captured PCM to
+ * {@link runDeviceVoiceTurn}; everything downstream is in-process JS + the
+ * fused FFI. No HTTP, no second voice path.
+ */
+import type { VoicePipelineEvents } from "../../services/voice/pipeline";
+import type { MtpTextRunner } from "../../services/voice/pipeline-impls";
+import type { TranscriptionAudio } from "../../services/voice/types";
+import type { VoicePreloadPredictor } from "../../services/voice/voice-preload-predictor";
+import type { CapacitorLlamaContext } from "./types";
+/** Turn exit reason, mirroring `LocalInferenceEngine.runVoiceTurn`. */
+export type VoiceTurnExitReason = "done" | "token-cap" | "cancelled";
+/**
+ * Structural view of `LocalInferenceEngine.runVoiceTurn` so the device bridge
+ * (and tests) depend on the call shape, not the concrete engine class.
+ */
+export interface DeviceVoiceEngine {
+    runVoiceTurn(audio: TranscriptionAudio, opts?: {
+        maxDraftTokens?: number;
+        maxGeneratedTokens?: number;
+        events?: VoicePipelineEvents;
+        textRunner?: MtpTextRunner;
+    }): Promise<VoiceTurnExitReason>;
+}
+export interface CapacitorTextRunnerOptions {
+    /** Default sampling temperature when the pipeline does not pin one. */
+    temperature?: number;
+    /** Default nucleus-sampling cutoff when the pipeline does not pin one. */
+    topP?: number;
+}
+/**
+ * Adapt a live `CapacitorLlamaContext` to the {@link MtpTextRunner} contract the
+ * fused voice pipeline drives.
+ *
+ * `hasDrafter()` is `false`: the on-device `CapacitorLlamaContext` does not yet
+ * expose a separate speculative-draft window, so the pipeline runs plain decode
+ * through the target verifier (`MtpTargetVerifier` splits the returned text into
+ * tokens when no per-delta verifier events arrive — its documented non-streaming
+ * path). When the on-device fork exposes draft/verify events this flips to
+ * `true` with no change to the call sites.
+ */
+export declare function createCapacitorMtpTextRunner(context: CapacitorLlamaContext, options?: CapacitorTextRunnerOptions): MtpTextRunner;
+export interface RunDeviceVoiceTurnArgs {
+    /** The engine that owns the armed voice bridge (`startVoice` + `armVoice`). */
+    engine: DeviceVoiceEngine;
+    /** The on-device text model context that backs the MTP target verifier. */
+    context: CapacitorLlamaContext;
+    /** Captured mic PCM for this turn. */
+    audio: TranscriptionAudio;
+    /** Pipeline events (first-audio, transcript, completion, cancel). */
+    events?: VoicePipelineEvents;
+    /** Upper bound on generated response tokens. */
+    maxGeneratedTokens?: number;
+    /** Default sampling overrides forwarded to the on-device text runner. */
+    generation?: CapacitorTextRunnerOptions;
+    /**
+     * Next-stage preload predictor (#8809 C5). On the mobile / Capacitor path the
+     * text-response model is owned by the {@link MemoryArbiter}; this turn's
+     * stages run sequentially (ASR → text → TTS), so the predictor warms the
+     * arbiter-managed text model the instant ASR finishes — during the ASR-stage
+     * page-trim window — collapsing the cold load off the post-ASR critical path.
+     * Wired to the pipeline's `onAsrComplete` event below. Omit to keep the turn
+     * unpredicted (the arbiter still loads the text model on first request).
+     */
+    preloadPredictor?: VoicePreloadPredictor;
+}
+/**
+ * Run one fused on-device voice turn. The native capture layer calls this with
+ * the captured mic audio; it builds the device text runner from the loaded
+ * `CapacitorLlamaContext` and drives `engine.runVoiceTurn`, so ASR + MTP +
+ * TTS all run in-process through the fused handle. Resolves with the turn's
+ * exit reason.
+ */
+export declare function runDeviceVoiceTurn(args: RunDeviceVoiceTurnArgs): Promise<VoiceTurnExitReason>;
+//# sourceMappingURL=voice-turn.d.ts.map

package/dist/adapters/capacitor-llama/voice-turn.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"voice-turn.d.ts","sourceRoot":"","sources":["../../../src/adapters/capacitor-llama/voice-turn.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAGH,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,+BAA+B,CAAC;AACzE,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,qCAAqC,CAAC;AACzE,OAAO,KAAK,EACX,kBAAkB,EAElB,MAAM,4BAA4B,CAAC;AACpC,OAAO,KAAK,EAAE,qBAAqB,EAAE,MAAM,8CAA8C,CAAC;AAC1F,OAAO,KAAK,EAAE,qBAAqB,EAAE,MAAM,SAAS,CAAC;AAErD,uEAAuE;AACvE,MAAM,MAAM,mBAAmB,GAAG,MAAM,GAAG,WAAW,GAAG,WAAW,CAAC;AAErE;;;GAGG;AACH,MAAM,WAAW,iBAAiB;IACjC,YAAY,CACX,KAAK,EAAE,kBAAkB,EACzB,IAAI,CAAC,EAAE;QACN,cAAc,CAAC,EAAE,MAAM,CAAC;QACxB,kBAAkB,CAAC,EAAE,MAAM,CAAC;QAC5B,MAAM,CAAC,EAAE,mBAAmB,CAAC;QAC7B,UAAU,CAAC,EAAE,aAAa,CAAC;KAC3B,GACC,OAAO,CAAC,mBAAmB,CAAC,CAAC;CAChC;AAED,MAAM,WAAW,0BAA0B;IAC1C,uEAAuE;IACvE,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,0EAA0E;IAC1E,IAAI,CAAC,EAAE,MAAM,CAAC;CACd;AAED;;;;;;;;;;GAUG;AACH,wBAAgB,4BAA4B,CAC3C,OAAO,EAAE,qBAAqB,EAC9B,OAAO,GAAE,0BAA+B,GACtC,aAAa,CAoCf;AAED,MAAM,WAAW,sBAAsB;IACtC,+EAA+E;IAC/E,MAAM,EAAE,iBAAiB,CAAC;IAC1B,2EAA2E;IAC3E,OAAO,EAAE,qBAAqB,CAAC;IAC/B,sCAAsC;IACtC,KAAK,EAAE,kBAAkB,CAAC;IAC1B,qEAAqE;IACrE,MAAM,CAAC,EAAE,mBAAmB,CAAC;IAC7B,gDAAgD;IAChD,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,yEAAyE;IACzE,UAAU,CAAC,EAAE,0BAA0B,CAAC;IACxC;;;;;;;;OAQG;IACH,gBAAgB,CAAC,EAAE,qBAAqB,CAAC;CACzC;AAED;;;;;;GAMG;AACH,wBAAgB,kBAAkB,CACjC,IAAI,EAAE,sBAAsB,GAC1B,OAAO,CAAC,mBAAmB,CAAC,CAa9B"}

package/dist/backends/apple-foundation.d.ts

@@ -0,0 +1,56 @@
+/**
+ * Apple Foundation Models adapter — iOS / iPadOS / macOS 26+.
+ *
+ * The Foundation Models framework exposes a managed on-device LLM that
+ * Apple ships with the OS when Apple Intelligence is enabled. We do not
+ * own its weights and cannot ship our own; this adapter is an
+ * **opportunistic fast-path**:
+ *
+ *   - If the device is iOS 26+ and reports availability via the Capacitor
+ *     `ComputerUse` bridge probe, we register a generate function that
+ *     calls into Apple's framework for short prompts.
+ *   - Otherwise we do nothing, and the existing llama-cpp-capacitor
+ *     (Gemma-VL / eliza-1 2b) path remains the active local-inference handler.
+ *
+ * Why this lives in plugin-local-inference (not plugin-capacitor-bridge):
+ *
+ *   - The plugin-capacitor-bridge package is a thin JS↔native shim. The
+ *     decision of *when* to use Apple's model vs. llama.cpp is an
+ *     inference-routing decision, which is plugin-local-inference's job.
+ *     Co-locating the adapter with the rest of the routing logic
+ *     (`backend-selector`, `runtime-dispatcher`, etc.) keeps the routing
+ *     surface inspectable in one place. The bridge only transports
+ *     bytes; it does not decide policy.
+ *
+ * Entitlement: requires `com.apple.developer.kernel.increased-memory-limit`
+ * and Apple Intelligence to be enabled by the user in Settings. See
+ * `docs/IOS_CONSTRAINTS.md`.
+ */
+import type { FoundationModelOptions, FoundationModelResult, IosComputerUseBridge } from "@elizaos/plugin-computeruse/mobile/ios-bridge";
+export interface AppleFoundationGenerateArgs {
+    readonly prompt: string;
+    readonly options?: FoundationModelOptions;
+}
+export interface AppleFoundationAdapter {
+    readonly name: "apple-foundation";
+    available(): boolean;
+    generate(args: AppleFoundationGenerateArgs): Promise<FoundationModelResult>;
+}
+/**
+ * Builds the adapter. The bridge getter is lazy so this module remains
+ * Node-importable for tests; in the Capacitor build the runtime resolves
+ * `Capacitor.Plugins.ComputerUse` and passes it in.
+ */
+export declare function createAppleFoundationAdapter(getBridge: () => IosComputerUseBridge | null): AppleFoundationAdapter;
+/**
+ * Register the adapter so the local-inference runtime picks it up
+ * opportunistically. Idempotent — subsequent calls overwrite the previous
+ * registration so a hot reload of the bridge swaps cleanly.
+ */
+export declare function registerAppleFoundationAdapter(adapter: AppleFoundationAdapter): void;
+export declare function getAppleFoundationAdapter(): AppleFoundationAdapter | null;
+/**
+ * Tests use this to ensure each spec starts from a known state.
+ */
+export declare function _resetAppleFoundationAdapterForTests(): void;
+//# sourceMappingURL=apple-foundation.d.ts.map

package/dist/backends/apple-foundation.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"apple-foundation.d.ts","sourceRoot":"","sources":["../../src/backends/apple-foundation.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG;AAEH,OAAO,KAAK,EACX,sBAAsB,EACtB,qBAAqB,EACrB,oBAAoB,EACpB,MAAM,+CAA+C,CAAC;AAEvD,MAAM,WAAW,2BAA2B;IAC3C,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,OAAO,CAAC,EAAE,sBAAsB,CAAC;CAC1C;AAED,MAAM,WAAW,sBAAsB;IACtC,QAAQ,CAAC,IAAI,EAAE,kBAAkB,CAAC;IAClC,SAAS,IAAI,OAAO,CAAC;IACrB,QAAQ,CAAC,IAAI,EAAE,2BAA2B,GAAG,OAAO,CAAC,qBAAqB,CAAC,CAAC;CAC5E;AAED;;;;GAIG;AACH,wBAAgB,4BAA4B,CAC3C,SAAS,EAAE,MAAM,oBAAoB,GAAG,IAAI,GAC1C,sBAAsB,CA+CxB;AAMD;;;;GAIG;AACH,wBAAgB,8BAA8B,CAC7C,OAAO,EAAE,sBAAsB,GAC7B,IAAI,CAEN;AAED,wBAAgB,yBAAyB,IAAI,sBAAsB,GAAG,IAAI,CAEzE;AAED;;GAEG;AACH,wBAAgB,oCAAoC,IAAI,IAAI,CAE3D"}

package/dist/index.d.ts

@@ -1,37 +1,8 @@
-import * as http from 'node:http';
-import { ContentValue } from '@elizaos/core';
-
-type LocalInferenceCommandIntent = "retry" | "resume" | "redownload" | "download" | "cancel" | "switch_smaller" | "status" | "use_cloud" | "use_local";
-interface LocalInferenceChatMetadata {
-    [key: string]: ContentValue;
-    intent?: LocalInferenceCommandIntent;
-    status: "missing" | "downloading" | "loading" | "failed" | "no_space" | "idle" | "ready" | "cancelled" | "routing";
-    modelId?: string | null;
-    activeModelId?: string | null;
-    provider?: string;
-    error?: string;
-    progress?: {
-        percent?: number;
-        receivedBytes: number;
-        totalBytes: number;
-        bytesPerSec?: number;
-        etaMs?: number | null;
-    };
-}
-interface LocalInferenceChatResult {
-    text: string;
-    localInference: LocalInferenceChatMetadata;
-}
-declare let activeModelState: {
-    modelId: string | null;
-    loadedAt: string | null;
-    status: "idle" | "loading" | "ready" | "error";
-    error?: string;
-};
-declare function getLocalInferenceActiveModelId(): string | undefined;
-declare function getLocalInferenceActiveSnapshot(): Promise<typeof activeModelState>;
-declare function getLocalInferenceChatStatus(intent?: LocalInferenceCommandIntent, error?: unknown): Promise<LocalInferenceChatResult>;
-declare function handleLocalInferenceChatCommand(intent: LocalInferenceCommandIntent, prompt: string): Promise<LocalInferenceChatResult>;
-declare function handleLocalInferenceRoutes(req: http.IncomingMessage, res: http.ServerResponse): Promise<boolean>;
-
-export { type LocalInferenceChatMetadata, type LocalInferenceChatResult, type LocalInferenceCommandIntent, getLocalInferenceActiveModelId, getLocalInferenceActiveSnapshot, getLocalInferenceChatStatus, handleLocalInferenceChatCommand, handleLocalInferenceRoutes };
+export { buildGenerateMediaHandler, detectMediaIntent, generateMediaAction, type MediaKind, } from "./actions/generate-media.js";
+export { extractSpeakerName, identifySpeakerAction, } from "./actions/identify-speaker.js";
+export { emitVoiceControl, startTranscriptionAction, stopTranscriptionAction, VOICE_CONTROL_STREAM, type VoiceControlCommand, type VoiceControlEvent, } from "./actions/transcription-control.js";
+export { getLocalInferenceActiveModelId, getLocalInferenceActiveSnapshot, getLocalInferenceChatStatus, handleLocalInferenceChatCommand, handleLocalInferenceRoutes, type LocalInferenceChatMetadata, type LocalInferenceChatResult, type LocalInferenceCommandIntent, } from "./local-inference-routes.js";
+export { createLocalInferenceModelHandlers, isLocalInferenceUnavailableError, LOCAL_INFERENCE_MODEL_TYPES, LOCAL_INFERENCE_PRIORITY, LOCAL_INFERENCE_PROVIDER_ID, LOCAL_INFERENCE_TEXT_MODEL_TYPES, LocalInferenceUnavailableError, type LocalInferenceUnavailableReason, localInferencePlugin, localInferencePlugin as default, } from "./provider.js";
+export { handleLocalInferenceTtsRoute, normalizeAudioBytes, sanitizeLocalInferenceSpeechText, sniffAudioContentType, } from "./routes/local-inference-tts-route.js";
+export { detectEmbeddingPreset, detectEmbeddingTier, EMBEDDING_PRESETS, type EmbeddingPreset, type EmbeddingTier, } from "./runtime/embedding-presets.js";
+//# sourceMappingURL=index.d.ts.map

package/dist/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAQA,OAAO,EACN,yBAAyB,EACzB,iBAAiB,EACjB,mBAAmB,EACnB,KAAK,SAAS,GACd,MAAM,6BAA6B,CAAC;AACrC,OAAO,EACN,kBAAkB,EAClB,qBAAqB,GACrB,MAAM,+BAA+B,CAAC;AACvC,OAAO,EACN,gBAAgB,EAChB,wBAAwB,EACxB,uBAAuB,EACvB,oBAAoB,EACpB,KAAK,mBAAmB,EACxB,KAAK,iBAAiB,GACtB,MAAM,oCAAoC,CAAC;AAC5C,OAAO,EACN,8BAA8B,EAC9B,+BAA+B,EAC/B,2BAA2B,EAC3B,+BAA+B,EAC/B,0BAA0B,EAC1B,KAAK,0BAA0B,EAC/B,KAAK,wBAAwB,EAC7B,KAAK,2BAA2B,GAChC,MAAM,6BAA6B,CAAC;AACrC,OAAO,EACN,iCAAiC,EACjC,gCAAgC,EAChC,2BAA2B,EAC3B,wBAAwB,EACxB,2BAA2B,EAC3B,gCAAgC,EAChC,8BAA8B,EAC9B,KAAK,+BAA+B,EACpC,oBAAoB,EACpB,oBAAoB,IAAI,OAAO,GAC/B,MAAM,eAAe,CAAC;AACvB,OAAO,EACN,4BAA4B,EAC5B,mBAAmB,EACnB,gCAAgC,EAChC,qBAAqB,GACrB,MAAM,uCAAuC,CAAC;AAE/C,OAAO,EACN,qBAAqB,EACrB,mBAAmB,EACnB,iBAAiB,EACjB,KAAK,eAAe,EACpB,KAAK,aAAa,GAClB,MAAM,gCAAgC,CAAC"}