@qvac/sdk 0.16.0 → 0.17.1

package/dist/schemas/common.js

@@ -28,8 +28,11 @@ import { modelRegistryListRequestSchema, modelRegistryListResponseSchema, modelR
 import { suspendRequestSchema, suspendResponseSchema } from './suspend.js';
 import { resumeRequestSchema, resumeResponseSchema } from './resume.js';
 import { stateRequestSchema, stateResponseSchema } from './state.js';
+import { getSystemResourcesRequestSchema, getSystemResourcesResponseSchema } from './system-resources.js';
 import { classifyRequestSchema, classifyResponseSchema } from './classification.js';
+import { audioGenStreamRequestSchema, audioGenStreamResponseSchema } from '../schemas/audio-gen.js';
 export const requestSchema = z.union([
+    audioGenStreamRequestSchema,
     heartbeatRequestSchema,
     loadModelRequestSchema,
     downloadAssetRequestSchema,
@@ -66,9 +69,11 @@ export const requestSchema = z.union([
     suspendRequestSchema,
     resumeRequestSchema,
     stateRequestSchema,
+    getSystemResourcesRequestSchema,
     classifyRequestSchema
 ]);
 export const responseSchema = z.discriminatedUnion('type', [
+    audioGenStreamResponseSchema,
     heartbeatResponseSchema,
     loadModelResponseSchema,
     downloadAssetResponseSchema,
@@ -109,6 +114,7 @@ export const responseSchema = z.discriminatedUnion('type', [
     suspendResponseSchema,
     resumeResponseSchema,
     stateResponseSchema,
+    getSystemResourcesResponseSchema,
     classifyResponseSchema
 ]);
 export const rpcOptionsSchema = z.object({

package/dist/schemas/completion-event.d.ts

@@ -1,12 +1,13 @@
 import { z } from 'zod';
-import { type Tool, type ToolCallWithCall } from './tools';
-import type { ToolDialect } from './completion-stream';
+import { type Tool, type ToolCallWithCall } from './tools.js';
+import type { ToolDialect } from './completion-stream.js';
 export declare const completionStatsSchema: z.ZodObject<{
     timeToFirstToken: z.ZodOptional<z.ZodNumber>;
     tokensPerSecond: z.ZodOptional<z.ZodNumber>;
     cacheTokens: z.ZodOptional<z.ZodNumber>;
     promptTokens: z.ZodOptional<z.ZodNumber>;
     generatedTokens: z.ZodOptional<z.ZodNumber>;
+    emittedTokens: z.ZodOptional<z.ZodNumber>;
     avgConcurrentSeq: z.ZodOptional<z.ZodNumber>;
     backendDevice: z.ZodOptional<z.ZodEnum<{
         gpu: "gpu";
@@ -62,6 +63,7 @@ export declare const statsEventSchema: z.ZodObject<{
         cacheTokens: z.ZodOptional<z.ZodNumber>;
         promptTokens: z.ZodOptional<z.ZodNumber>;
         generatedTokens: z.ZodOptional<z.ZodNumber>;
+        emittedTokens: z.ZodOptional<z.ZodNumber>;
         avgConcurrentSeq: z.ZodOptional<z.ZodNumber>;
         backendDevice: z.ZodOptional<z.ZodEnum<{
             gpu: "gpu";
@@ -143,6 +145,7 @@ export declare const completionEventSchema: z.ZodUnion<readonly [z.ZodObject<{
         cacheTokens: z.ZodOptional<z.ZodNumber>;
         promptTokens: z.ZodOptional<z.ZodNumber>;
         generatedTokens: z.ZodOptional<z.ZodNumber>;
+        emittedTokens: z.ZodOptional<z.ZodNumber>;
         avgConcurrentSeq: z.ZodOptional<z.ZodNumber>;
         backendDevice: z.ZodOptional<z.ZodEnum<{
             gpu: "gpu";

package/dist/schemas/completion-event.js

@@ -5,7 +5,10 @@ export const completionStatsSchema = z.object({
     tokensPerSecond: z.number().optional(),
     cacheTokens: z.number().optional(),
     promptTokens: z.number().optional(),
+    // Decode count (`llama_perf` `n_eval`) — length / KV-cache budget decisions.
     generatedTokens: z.number().optional(),
+    // Non-empty addon stream pieces — prefer for OpenAI usage when present.
+    emittedTokens: z.number().optional(),
     avgConcurrentSeq: z.number().optional(),
     backendDevice: z.enum(['cpu', 'gpu']).optional()
 });

package/dist/schemas/completion-stream.d.ts

@@ -1,5 +1,5 @@
 import { z } from 'zod';
-export { completionStatsSchema, type CompletionStats } from './completion-event';
+export { completionStatsSchema, type CompletionStats } from './completion-event.js';
 /**
  * Tool-call output dialect. Auto-detected from the model name; pass via
  * `completion({ toolDialect })` to override.
@@ -11,6 +11,8 @@ export { completionStatsSchema, type CompletionStats } from './completion-event'
  * - `"harmony"`:  `<|channel|>commentary to=functions.get_weather <|constrain|>json<|message|>{"city":"Tokyo"}<|call|>`
  * - `"qwen35"`:   `<tool_call><function=NAME><parameter=KEY>VALUE</parameter></function></tool_call>`
  * - `"gemma4"`:   `<|tool_call>call:NAME{key:<|"|>val<|"|>,...}<tool_call|>`
+ * - `"dsml"`:     `<｜DSML｜tool_calls><｜DSML｜invoke name="NAME"><｜DSML｜parameter name="KEY" string="true">VALUE</｜DSML｜parameter></｜DSML｜invoke></｜DSML｜tool_calls>`
+ *                 (DeepSeek Markup Language, emitted by DeepSeek V3.2 / V4)
  */
 export declare const toolDialectSchema: z.ZodEnum<{
     hermes: "hermes";
@@ -19,6 +21,7 @@ export declare const toolDialectSchema: z.ZodEnum<{
     harmony: "harmony";
     qwen35: "qwen35";
     gemma4: "gemma4";
+    dsml: "dsml";
 }>;
 export declare const attachmentSchema: z.ZodObject<{
     path: z.ZodString;
@@ -112,6 +115,7 @@ export declare const completionClientParamsSchema: z.ZodObject<{
         harmony: "harmony";
         qwen35: "qwen35";
         gemma4: "gemma4";
+        dsml: "dsml";
     }>>;
     responseFormat: z.ZodOptional<z.ZodDiscriminatedUnion<[z.ZodObject<{
         type: z.ZodLiteral<"text">;
@@ -181,6 +185,7 @@ export declare const completionStreamRequestSchema: z.ZodObject<{
         harmony: "harmony";
         qwen35: "qwen35";
         gemma4: "gemma4";
+        dsml: "dsml";
     }>>;
     responseFormat: z.ZodOptional<z.ZodDiscriminatedUnion<[z.ZodObject<{
         type: z.ZodLiteral<"text">;
@@ -243,6 +248,7 @@ export declare const completionStreamResponseSchema: z.ZodObject<{
             cacheTokens: z.ZodOptional<z.ZodNumber>;
             promptTokens: z.ZodOptional<z.ZodNumber>;
             generatedTokens: z.ZodOptional<z.ZodNumber>;
+            emittedTokens: z.ZodOptional<z.ZodNumber>;
             avgConcurrentSeq: z.ZodOptional<z.ZodNumber>;
             backendDevice: z.ZodOptional<z.ZodEnum<{
                 gpu: "gpu";
@@ -326,6 +332,7 @@ export declare const completionOrchestrateRequestSchema: z.ZodObject<{
         harmony: "harmony";
         qwen35: "qwen35";
         gemma4: "gemma4";
+        dsml: "dsml";
     }>>;
     responseFormat: z.ZodOptional<z.ZodDiscriminatedUnion<[z.ZodObject<{
         type: z.ZodLiteral<"text">;
@@ -406,6 +413,7 @@ export declare const completionOrchestrateResponseSchema: z.ZodObject<{
             cacheTokens: z.ZodOptional<z.ZodNumber>;
             promptTokens: z.ZodOptional<z.ZodNumber>;
             generatedTokens: z.ZodOptional<z.ZodNumber>;
+            emittedTokens: z.ZodOptional<z.ZodNumber>;
             avgConcurrentSeq: z.ZodOptional<z.ZodNumber>;
             backendDevice: z.ZodOptional<z.ZodEnum<{
                 gpu: "gpu";

package/dist/schemas/completion-stream.js

@@ -14,6 +14,8 @@ export { completionStatsSchema } from './completion-event.js';
  * - `"harmony"`:  `<|channel|>commentary to=functions.get_weather <|constrain|>json<|message|>{"city":"Tokyo"}<|call|>`
  * - `"qwen35"`:   `<tool_call><function=NAME><parameter=KEY>VALUE</parameter></function></tool_call>`
  * - `"gemma4"`:   `<|tool_call>call:NAME{key:<|"|>val<|"|>,...}<tool_call|>`
+ * - `"dsml"`:     `<｜DSML｜tool_calls><｜DSML｜invoke name="NAME"><｜DSML｜parameter name="KEY" string="true">VALUE</｜DSML｜parameter></｜DSML｜invoke></｜DSML｜tool_calls>`
+ *                 (DeepSeek Markup Language, emitted by DeepSeek V3.2 / V4)
  */
 export const toolDialectSchema = z.enum([
     'hermes',
@@ -21,7 +23,8 @@ export const toolDialectSchema = z.enum([
     'json',
     'harmony',
     'qwen35',
-    'gemma4'
+    'gemma4',
+    'dsml'
 ]);
 export const attachmentSchema = z.object({
     path: z
@@ -61,7 +64,7 @@ export const generationParamsSchema = z
     remove_thinking_from_context: z
         .boolean()
         .optional()
-        .describe('When the model emits a reasoning block during generation (e.g. `<think>...</think>` for the Qwen3 family, `<|channel>thought ... <channel|>` for Gemma 4), drop those tokens from the KV cache at end-of-generation so subsequent turns do not accumulate reasoning history. Defaults to `false`. No-op for models without a recognised reasoning channel. Throws on models with recurrent memory (SSM / hybrid SSM such as Qwen3.5), where the cache edit is unsupported.')
+        .describe('When the model emits a reasoning block during generation (e.g. `<think>...</think>` for the Qwen3 family, `<|channel>thought ... <channel|>` for Gemma 4), drop those tokens from the KV cache at end-of-generation so subsequent turns do not accumulate reasoning history. Defaults to `false`, except the Qwen3 reasoning family (Qwen3, Qwen3.5, Qwen3.6, including MoE variants), which defaults to `true`. No-op for models without a recognised reasoning channel. Supported on recurrent / hybrid-SSM models (e.g. Qwen3.5) via a state snapshot and replay when the reasoning close marker is a single token; on such a model with a multi-token close marker, enabling this fails with an error.')
 })
     .strict();
 const jsonSchemaObjectSchema = z.record(z.string(), z.unknown());

package/dist/schemas/constants-registry.d.ts

@@ -35,6 +35,7 @@ export declare const constantsRegistry: {
         readonly parakeetTranscription: "parakeet-transcription";
         readonly ggmlOcr: "ggml-ocr";
         readonly sdcppGeneration: "sdcpp-generation";
+        readonly audiogenGgml: "audiogen-ggml";
         readonly ggmlVla: "ggml-vla";
         readonly ggmlClassification: "ggml-classification";
     }>;
@@ -57,6 +58,7 @@ export declare const constantsRegistry: {
         readonly TTS: "@qvac/sdk/tts-ggml/plugin";
         readonly OCR: "@qvac/sdk/ggml-ocr/plugin";
         readonly DIFFUSION: "@qvac/sdk/sdcpp-generation/plugin";
+        readonly AUDIOGEN: "@qvac/sdk/audiogen-ggml/plugin";
         readonly VLA: "@qvac/sdk/ggml-vla/plugin";
         readonly CLASSIFICATION: "@qvac/sdk/ggml-classification/plugin";
     }>;

package/dist/schemas/constants-registry.js

@@ -2,7 +2,7 @@ import { z } from 'zod';
 import { ModelType } from './model-types.js';
 import { TOOLS_MODE } from './tools.js';
 import { VERBOSITY } from './llamacpp-config.js';
-import { PLUGIN_LLM, PLUGIN_EMBEDDING, PLUGIN_WHISPER, PLUGIN_BCI, PLUGIN_NMT, PLUGIN_TTS, PLUGIN_OCR, PLUGIN_DIFFUSION, PLUGIN_VLA, PLUGIN_CLASSIFICATION } from './plugin.js';
+import { PLUGIN_LLM, PLUGIN_EMBEDDING, PLUGIN_WHISPER, PLUGIN_BCI, PLUGIN_NMT, PLUGIN_TTS, PLUGIN_OCR, PLUGIN_DIFFUSION, PLUGIN_AUDIOGEN, PLUGIN_VLA, PLUGIN_CLASSIFICATION } from './plugin.js';
 import { SUPPORTED_AUDIO_FORMATS } from '../constants/audio.js';
 /**
  * Every public constant `enum` from index.ts that downstream (non-JS)
@@ -41,6 +41,7 @@ export const constantsRegistry = {
         TTS: PLUGIN_TTS,
         OCR: PLUGIN_OCR,
         DIFFUSION: PLUGIN_DIFFUSION,
+        AUDIOGEN: PLUGIN_AUDIOGEN,
         VLA: PLUGIN_VLA,
         CLASSIFICATION: PLUGIN_CLASSIFICATION
     }),

package/dist/schemas/download-asset.d.ts

@@ -1,5 +1,5 @@
 import { z } from 'zod';
-import type { ModelProgressUpdate } from './load-model';
+import type { ModelProgressUpdate } from './load-model.js';
 export declare const downloadAssetOptionsSchema: z.ZodPipe<z.ZodObject<{
     assetSrc: z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
         src: z.ZodString;
@@ -15,6 +15,7 @@ export declare const downloadAssetOptionsSchema: z.ZodPipe<z.ZodObject<{
         addon: z.ZodOptional<z.ZodUnion<readonly [z.ZodEnum<{
             embeddings: "embeddings";
             diffusion: "diffusion";
+            audiogen: "audiogen";
             tts: "tts";
             ocr: "ocr";
             vla: "vla";
@@ -28,6 +29,7 @@ export declare const downloadAssetOptionsSchema: z.ZodPipe<z.ZodObject<{
             "parakeet-transcription": "parakeet-transcription";
             "ggml-ocr": "ggml-ocr";
             "sdcpp-generation": "sdcpp-generation";
+            "audiogen-ggml": "audiogen-ggml";
             "ggml-vla": "ggml-vla";
             "ggml-classification": "ggml-classification";
             llm: "llm";
@@ -52,7 +54,7 @@ export declare const downloadAssetOptionsSchema: z.ZodPipe<z.ZodObject<{
         engine?: string | undefined;
         expectedSize?: number | undefined;
         sha256Checksum?: string | undefined;
-        addon?: "embeddings" | "diffusion" | "tts" | "ocr" | "vla" | "llamacpp-completion" | "whispercpp-transcription" | "bci-whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "tts-ggml" | "parakeet-transcription" | "ggml-ocr" | "sdcpp-generation" | "ggml-vla" | "ggml-classification" | "llm" | "whisper" | "bci" | "nmt" | "parakeet" | "classification" | "vad" | undefined;
+        addon?: "embeddings" | "diffusion" | "audiogen" | "tts" | "ocr" | "vla" | "llamacpp-completion" | "whispercpp-transcription" | "bci-whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "tts-ggml" | "parakeet-transcription" | "ggml-ocr" | "sdcpp-generation" | "audiogen-ggml" | "ggml-vla" | "ggml-classification" | "llm" | "whisper" | "bci" | "nmt" | "parakeet" | "classification" | "vad" | undefined;
     };
 }, {
     assetSrc: string | {
@@ -66,7 +68,7 @@ export declare const downloadAssetOptionsSchema: z.ZodPipe<z.ZodObject<{
         engine?: string | undefined;
         expectedSize?: number | undefined;
         sha256Checksum?: string | undefined;
-        addon?: "embeddings" | "diffusion" | "tts" | "ocr" | "vla" | "llamacpp-completion" | "whispercpp-transcription" | "bci-whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "tts-ggml" | "parakeet-transcription" | "ggml-ocr" | "sdcpp-generation" | "ggml-vla" | "ggml-classification" | "llm" | "whisper" | "bci" | "nmt" | "parakeet" | "classification" | "vad" | undefined;
+        addon?: "embeddings" | "diffusion" | "audiogen" | "tts" | "ocr" | "vla" | "llamacpp-completion" | "whispercpp-transcription" | "bci-whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "tts-ggml" | "parakeet-transcription" | "ggml-ocr" | "sdcpp-generation" | "audiogen-ggml" | "ggml-vla" | "ggml-classification" | "llm" | "whisper" | "bci" | "nmt" | "parakeet" | "classification" | "vad" | undefined;
     };
     seed?: boolean | undefined;
 }>>;
@@ -85,6 +87,7 @@ export declare const downloadAssetOptionsToRequestSchema: z.ZodPipe<z.ZodObject<
         addon: z.ZodOptional<z.ZodUnion<readonly [z.ZodEnum<{
             embeddings: "embeddings";
             diffusion: "diffusion";
+            audiogen: "audiogen";
             tts: "tts";
             ocr: "ocr";
             vla: "vla";
@@ -98,6 +101,7 @@ export declare const downloadAssetOptionsToRequestSchema: z.ZodPipe<z.ZodObject<
             "parakeet-transcription": "parakeet-transcription";
             "ggml-ocr": "ggml-ocr";
             "sdcpp-generation": "sdcpp-generation";
+            "audiogen-ggml": "audiogen-ggml";
             "ggml-vla": "ggml-vla";
             "ggml-classification": "ggml-classification";
             llm: "llm";
@@ -130,7 +134,7 @@ export declare const downloadAssetOptionsToRequestSchema: z.ZodPipe<z.ZodObject<
         engine?: string | undefined;
         expectedSize?: number | undefined;
         sha256Checksum?: string | undefined;
-        addon?: "embeddings" | "diffusion" | "tts" | "ocr" | "vla" | "llamacpp-completion" | "whispercpp-transcription" | "bci-whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "tts-ggml" | "parakeet-transcription" | "ggml-ocr" | "sdcpp-generation" | "ggml-vla" | "ggml-classification" | "llm" | "whisper" | "bci" | "nmt" | "parakeet" | "classification" | "vad" | undefined;
+        addon?: "embeddings" | "diffusion" | "audiogen" | "tts" | "ocr" | "vla" | "llamacpp-completion" | "whispercpp-transcription" | "bci-whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "tts-ggml" | "parakeet-transcription" | "ggml-ocr" | "sdcpp-generation" | "audiogen-ggml" | "ggml-vla" | "ggml-classification" | "llm" | "whisper" | "bci" | "nmt" | "parakeet" | "classification" | "vad" | undefined;
     };
     seed?: boolean | undefined;
     onProgress?: unknown;

package/dist/schemas/engine-addon-map.d.ts

@@ -1,4 +1,4 @@
-import { type ModelRegistryEngine, type ModelRegistryEntryAddon } from './registry';
+import { type ModelRegistryEngine, type ModelRegistryEntryAddon } from './registry.js';
 export declare const ENGINE_TO_ADDON: {
     readonly "llamacpp-completion": "llm";
     readonly "whispercpp-transcription": "whisper";
@@ -10,6 +10,7 @@ export declare const ENGINE_TO_ADDON: {
     readonly "ggml-ocr": "ocr";
     readonly "parakeet-transcription": "parakeet";
     readonly "sdcpp-generation": "diffusion";
+    readonly "audiogen-ggml": "audiogen";
     readonly "ggml-vla": "vla";
     readonly "ggml-classification": "classification";
     readonly 'onnx-vad': "vad";

package/dist/schemas/engine-addon-map.js

@@ -1,5 +1,5 @@
 import { ModelType } from './model-types.js';
-import { ADDON_BCI, ADDON_CLASSIFICATION, ADDON_DIFFUSION, ADDON_EMBEDDING, ADDON_LLM, ADDON_NMT, ADDON_OCR, ADDON_PARAKEET, ADDON_TTS, ADDON_VLA, ADDON_WHISPER } from './plugin.js';
+import { ADDON_BCI, ADDON_CLASSIFICATION, ADDON_DIFFUSION, ADDON_AUDIOGEN, ADDON_EMBEDDING, ADDON_LLM, ADDON_NMT, ADDON_OCR, ADDON_PARAKEET, ADDON_TTS, ADDON_VLA, ADDON_WHISPER } from './plugin.js';
 import { modelRegistryEngineSchema } from './registry.js';
 // Canonical engine → addon mapping (exhaustive). `as const` preserves
 // per-key literals so the addon can be derived from the engine at the type level.
@@ -14,6 +14,7 @@ export const ENGINE_TO_ADDON = {
     [ModelType.ggmlOcr]: 'ocr',
     [ModelType.parakeetTranscription]: 'parakeet',
     [ModelType.sdcppGeneration]: 'diffusion',
+    [ModelType.audiogenGgml]: 'audiogen',
     [ModelType.ggmlVla]: 'vla',
     [ModelType.ggmlClassification]: 'classification',
     'onnx-vad': 'vad'
@@ -48,6 +49,9 @@ export const LEGACY_ENGINE_TO_CANONICAL = {
     ocr: ModelType.ggmlOcr,
     [ADDON_DIFFUSION]: ModelType.sdcppGeneration,
     diffusion: ModelType.sdcppGeneration,
+    [ADDON_AUDIOGEN]: ModelType.audiogenGgml,
+    '@qvac/audiogen': ModelType.audiogenGgml,
+    audiogen: ModelType.audiogenGgml,
     [ADDON_VLA]: ModelType.ggmlVla,
     vla: ModelType.ggmlVla,
     [ADDON_CLASSIFICATION]: ModelType.ggmlClassification,

package/dist/schemas/get-loaded-model-info.d.ts

@@ -23,6 +23,7 @@ export declare const loadedModelInfoSchema: z.ZodDiscriminatedUnion<[z.ZodObject
         harmony: "harmony";
         qwen35: "qwen35";
         gemma4: "gemma4";
+        dsml: "dsml";
     }>>;
 }, z.core.$strip>, z.ZodObject<{
     modelId: z.ZodString;
@@ -51,6 +52,7 @@ export declare const getLoadedModelInfoResponseSchema: z.ZodObject<{
             harmony: "harmony";
             qwen35: "qwen35";
             gemma4: "gemma4";
+            dsml: "dsml";
         }>>;
     }, z.core.$strip>, z.ZodObject<{
         modelId: z.ZodString;

package/dist/schemas/get-model-info.d.ts

@@ -37,6 +37,7 @@ export declare const modelInfoSchema: z.ZodObject<{
     addon: z.ZodEnum<{
         embeddings: "embeddings";
         diffusion: "diffusion";
+        audiogen: "audiogen";
         tts: "tts";
         ocr: "ocr";
         vla: "vla";
@@ -87,6 +88,7 @@ export declare const getModelInfoResponseSchema: z.ZodObject<{
         addon: z.ZodEnum<{
             embeddings: "embeddings";
             diffusion: "diffusion";
+            audiogen: "audiogen";
             tts: "tts";
             ocr: "ocr";
             vla: "vla";

package/dist/schemas/get-model-info.js

@@ -64,6 +64,7 @@ export const modelInfoSchema = z.object({
         'tts',
         'ocr',
         'diffusion',
+        'audiogen',
         'vla',
         'classification',
         'other'

package/dist/schemas/index.d.ts

@@ -1,53 +1,55 @@
-export * from './archive';
-export * from './cancel';
-export * from './batch-completion-stream';
-export * from './completion-stream';
-export * from './completion-event';
-export { toolSchema, toolCallSchema, toolCallErrorSchema, TOOLS_MODE, type Tool, type ToolCall, type ToolCallError, type ToolCallWithCall, type ToolsMode } from './tools';
-export * from './delegate';
-export * from './delete-cache';
-export * from './download-asset';
-export * from './embed';
-export * from './finetune';
-export * from './load-model';
-export * from './reload-config';
-export * from './logging-stream';
-export * from './provide';
-export * from './stop-provide';
-export * from './unload-model';
-export * from './heartbeat';
-export * from './common';
-export * from './transcription';
-export * from './bci';
-export * from './bci-config';
-export * from './translate';
-export * from './translation-config';
-export * from './llamacpp-config';
-export * from './transcription-config';
-export * from './text-to-speech';
-export * from './error';
-export * from './rag';
-export * from './ocr';
-export * from './sdcpp-config';
-export * from './vla';
-export * from './classification';
-export * from './shard';
-export * from './suspend';
-export * from './resume';
-export * from './state';
-export { SDK_CLIENT_ERROR_CODES } from './sdk-errors-client';
-export { SDK_SERVER_ERROR_CODES } from './sdk-errors-server';
-export { REGISTRY_ERROR_CODES } from './sdk-errors-registry';
+export * from './archive.js';
+export * from './cancel.js';
+export * from './batch-completion-stream.js';
+export * from './completion-stream.js';
+export * from './completion-event.js';
+export { toolSchema, toolCallSchema, toolCallErrorSchema, TOOLS_MODE, type Tool, type ToolCall, type ToolCallError, type ToolCallWithCall, type ToolsMode } from './tools.js';
+export * from './delegate.js';
+export * from './delete-cache.js';
+export * from './download-asset.js';
+export * from './embed.js';
+export * from './finetune.js';
+export * from './load-model.js';
+export * from './reload-config.js';
+export * from './logging-stream.js';
+export * from './provide.js';
+export * from './stop-provide.js';
+export * from './unload-model.js';
+export * from './heartbeat.js';
+export * from './common.js';
+export * from './transcription.js';
+export * from './bci.js';
+export * from './bci-config.js';
+export * from './translate.js';
+export * from './translation-config.js';
+export * from './llamacpp-config.js';
+export * from './transcription-config.js';
+export * from './text-to-speech.js';
+export * from './error.js';
+export * from './rag.js';
+export * from './ocr.js';
+export * from './sdcpp-config.js';
+export * from './vla.js';
+export * from './classification.js';
+export * from './audio-gen.js';
+export * from './shard.js';
+export * from './suspend.js';
+export * from './resume.js';
+export * from './state.js';
+export * from './system-resources.js';
+export { SDK_CLIENT_ERROR_CODES } from './sdk-errors-client.js';
+export { SDK_SERVER_ERROR_CODES } from './sdk-errors-server.js';
+export { REGISTRY_ERROR_CODES } from './sdk-errors-registry.js';
 export { ERR_CODES as RAG_ERROR_CODES } from '@qvac/rag/errors';
-export { qvacConfigSchema, deviceMatchSchema, deviceConfigDefaultsSchema, devicePatternSchema, type QvacConfig, type DeviceMatch, type DeviceConfigDefaults, type DevicePattern } from './sdk-config';
-export { PROFILING_KEY, PROFILING_TRAILER_KEY, DELEGATION_BREAKDOWN_KEY, OPERATION_EVENT_KEY, MODEL_EXECUTION_KEY, profilerModeSchema, serverBreakdownSchema, delegationBreakdownSchema, operationEventSchema, profilingRequestMetaSchema, profilingResponseMetaSchema, perCallProfilingSchema, type ProfilerMode, type ServerBreakdown, type DelegationBreakdown, type OperationEvent, type ProfilingRequestMeta, type ProfilingResponseMeta, type PerCallProfiling } from './profiling';
-export { runtimeContextSchema, type RuntimeContext } from './runtime-context';
-export * from './get-model-info';
-export * from './get-loaded-model-info';
-export * from './model-src-utils';
-export * from './json-schema';
-export { type McpClient, type McpClientInput } from './mcp-adapter';
-export { PUBLIC_MODEL_TYPES as MODEL_TYPES, ModelType, type CanonicalModelType, type ModelTypeInput, normalizeModelType, isCanonicalModelType, isModelTypeAlias } from './model-types';
-export * from './plugin';
-export * from './registry';
+export { qvacConfigSchema, deviceMatchSchema, deviceConfigDefaultsSchema, devicePatternSchema, type QvacConfig, type DeviceMatch, type DeviceConfigDefaults, type DevicePattern } from './sdk-config.js';
+export { PROFILING_KEY, PROFILING_TRAILER_KEY, DELEGATION_BREAKDOWN_KEY, OPERATION_EVENT_KEY, MODEL_EXECUTION_KEY, BACKEND_DIAGNOSTICS_KEY, profilerModeSchema, profilerGPUResourceGaugeSchema, profilerResourceGaugeSchema, serverBreakdownSchema, delegationBreakdownSchema, operationEventSchema, profilingRequestMetaSchema, profilingResponseMetaSchema, perCallProfilingSchema, type ProfilerMode, type ServerBreakdown, type DelegationBreakdown, type OperationEvent, type ProfilingRequestMeta, type ProfilingResponseMeta, type PerCallProfiling, type ProfilerGPUResourceGauge, type ProfilerResourceGauge } from './profiling.js';
+export { runtimeContextSchema, type RuntimeContext } from './runtime-context.js';
+export * from './get-model-info.js';
+export * from './get-loaded-model-info.js';
+export * from './model-src-utils.js';
+export * from './json-schema.js';
+export { type McpClient, type McpClientInput } from './mcp-adapter.js';
+export { PUBLIC_MODEL_TYPES as MODEL_TYPES, ModelType, type CanonicalModelType, type ModelTypeInput, normalizeModelType, isCanonicalModelType, isModelTypeAlias } from './model-types.js';
+export * from './plugin.js';
+export * from './registry.js';
 //# sourceMappingURL=index.d.ts.map

package/dist/schemas/index.js

@@ -32,16 +32,18 @@ export * from './ocr.js';
 export * from './sdcpp-config.js';
 export * from './vla.js';
 export * from './classification.js';
+export * from './audio-gen.js';
 export * from './shard.js';
 export * from './suspend.js';
 export * from './resume.js';
 export * from './state.js';
+export * from './system-resources.js';
 export { SDK_CLIENT_ERROR_CODES } from './sdk-errors-client.js';
 export { SDK_SERVER_ERROR_CODES } from './sdk-errors-server.js';
 export { REGISTRY_ERROR_CODES } from './sdk-errors-registry.js';
-export { ERR_CODES as RAG_ERROR_CODES } from '@qvac/rag/errors.js';
+export { ERR_CODES as RAG_ERROR_CODES } from '@qvac/rag/errors';
 export { qvacConfigSchema, deviceMatchSchema, deviceConfigDefaultsSchema, devicePatternSchema } from './sdk-config.js';
-export { PROFILING_KEY, PROFILING_TRAILER_KEY, DELEGATION_BREAKDOWN_KEY, OPERATION_EVENT_KEY, MODEL_EXECUTION_KEY, profilerModeSchema, serverBreakdownSchema, delegationBreakdownSchema, operationEventSchema, profilingRequestMetaSchema, profilingResponseMetaSchema, perCallProfilingSchema } from './profiling.js';
+export { PROFILING_KEY, PROFILING_TRAILER_KEY, DELEGATION_BREAKDOWN_KEY, OPERATION_EVENT_KEY, MODEL_EXECUTION_KEY, BACKEND_DIAGNOSTICS_KEY, profilerModeSchema, profilerGPUResourceGaugeSchema, profilerResourceGaugeSchema, serverBreakdownSchema, delegationBreakdownSchema, operationEventSchema, profilingRequestMetaSchema, profilingResponseMetaSchema, perCallProfilingSchema } from './profiling.js';
 export { runtimeContextSchema } from './runtime-context.js';
 export * from './get-model-info.js';
 export * from './get-loaded-model-info.js';

package/dist/schemas/llamacpp-config.d.ts

@@ -68,6 +68,7 @@ export declare const llmConfigBaseSchema: z.ZodObject<{
         addon: z.ZodOptional<z.ZodUnion<readonly [z.ZodEnum<{
             embeddings: "embeddings";
             diffusion: "diffusion";
+            audiogen: "audiogen";
             tts: "tts";
             ocr: "ocr";
             vla: "vla";
@@ -81,6 +82,7 @@ export declare const llmConfigBaseSchema: z.ZodObject<{
             "parakeet-transcription": "parakeet-transcription";
             "ggml-ocr": "ggml-ocr";
             "sdcpp-generation": "sdcpp-generation";
+            "audiogen-ggml": "audiogen-ggml";
             "ggml-vla": "ggml-vla";
             "ggml-classification": "ggml-classification";
             llm: "llm";
@@ -163,6 +165,7 @@ export declare const llmConfigSchema: z.ZodPipe<z.ZodObject<{
         addon: z.ZodOptional<z.ZodUnion<readonly [z.ZodEnum<{
             embeddings: "embeddings";
             diffusion: "diffusion";
+            audiogen: "audiogen";
             tts: "tts";
             ocr: "ocr";
             vla: "vla";
@@ -176,6 +179,7 @@ export declare const llmConfigSchema: z.ZodPipe<z.ZodObject<{
             "parakeet-transcription": "parakeet-transcription";
             "ggml-ocr": "ggml-ocr";
             "sdcpp-generation": "sdcpp-generation";
+            "audiogen-ggml": "audiogen-ggml";
             "ggml-vla": "ggml-vla";
             "ggml-classification": "ggml-classification";
             llm: "llm";
@@ -231,7 +235,7 @@ export declare const llmConfigSchema: z.ZodPipe<z.ZodObject<{
         engine?: string | undefined;
         expectedSize?: number | undefined;
         sha256Checksum?: string | undefined;
-        addon?: "embeddings" | "diffusion" | "tts" | "ocr" | "vla" | "llamacpp-completion" | "whispercpp-transcription" | "bci-whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "tts-ggml" | "parakeet-transcription" | "ggml-ocr" | "sdcpp-generation" | "ggml-vla" | "ggml-classification" | "llm" | "whisper" | "bci" | "nmt" | "parakeet" | "classification" | "vad" | undefined;
+        addon?: "embeddings" | "diffusion" | "audiogen" | "tts" | "ocr" | "vla" | "llamacpp-completion" | "whispercpp-transcription" | "bci-whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "tts-ggml" | "parakeet-transcription" | "ggml-ocr" | "sdcpp-generation" | "audiogen-ggml" | "ggml-vla" | "ggml-classification" | "llm" | "whisper" | "bci" | "nmt" | "parakeet" | "classification" | "vad" | undefined;
     } | undefined;
     image_tile_mode: "disabled" | "batched" | "sequential" | undefined;
     'mmproj-use-gpu'?: boolean | undefined;
@@ -274,7 +278,7 @@ export declare const llmConfigSchema: z.ZodPipe<z.ZodObject<{
         engine?: string | undefined;
         expectedSize?: number | undefined;
         sha256Checksum?: string | undefined;
-        addon?: "embeddings" | "diffusion" | "tts" | "ocr" | "vla" | "llamacpp-completion" | "whispercpp-transcription" | "bci-whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "tts-ggml" | "parakeet-transcription" | "ggml-ocr" | "sdcpp-generation" | "ggml-vla" | "ggml-classification" | "llm" | "whisper" | "bci" | "nmt" | "parakeet" | "classification" | "vad" | undefined;
+        addon?: "embeddings" | "diffusion" | "audiogen" | "tts" | "ocr" | "vla" | "llamacpp-completion" | "whispercpp-transcription" | "bci-whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "tts-ggml" | "parakeet-transcription" | "ggml-ocr" | "sdcpp-generation" | "audiogen-ggml" | "ggml-vla" | "ggml-classification" | "llm" | "whisper" | "bci" | "nmt" | "parakeet" | "classification" | "vad" | undefined;
     } | undefined;
     image_tile_mode?: "disabled" | "batched" | "sequential" | undefined;
     'mmproj-use-gpu'?: boolean | undefined;