@qvac/sdk 0.4.0 → 0.5.0

package/dist/schemas/common.d.ts

@@ -1,7 +1,7 @@
 import { z } from "zod";
 export declare const requestSchema: z.ZodUnion<readonly [z.ZodObject<{
     type: z.ZodLiteral<"ping">;
-}, z.core.$strip>, z.ZodPipe<z.ZodDiscriminatedUnion<[z.ZodObject<{
+}, z.core.$strip>, z.ZodUnion<readonly [z.ZodPipe<z.ZodDiscriminatedUnion<[z.ZodObject<{
     type: z.ZodLiteral<"loadModel">;
     modelSrc: z.ZodString;
     modelName: z.ZodOptional<z.ZodString>;
@@ -29,7 +29,7 @@ export declare const requestSchema: z.ZodUnion<readonly [z.ZodObject<{
         predict: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<-1>, z.ZodLiteral<-2>, z.ZodNumber]>>;
         system_prompt: z.ZodDefault<z.ZodString>;
         no_mmap: z.ZodOptional<z.ZodBoolean>;
-        verbosity: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<-1>, z.ZodLiteral<0>, z.ZodLiteral<1>]>>;
+        verbosity: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<0>, z.ZodLiteral<1>, z.ZodLiteral<2>, z.ZodLiteral<3>]>>;
         presence_penalty: z.ZodOptional<z.ZodNumber>;
         frequency_penalty: z.ZodOptional<z.ZodNumber>;
         repeat_penalty: z.ZodOptional<z.ZodNumber>;
@@ -101,7 +101,10 @@ export declare const requestSchema: z.ZodUnion<readonly [z.ZodObject<{
             speech_pad_ms: z.ZodOptional<z.ZodNumber>;
             samples_overlap: z.ZodOptional<z.ZodNumber>;
         }, z.core.$strip>>;
-        audio_format: z.ZodOptional<z.ZodString>;
+        audio_format: z.ZodOptional<z.ZodEnum<{
+            f32le: "f32le";
+            s16le: "s16le";
+        }>>;
         contextParams: z.ZodOptional<z.ZodObject<{
             model: z.ZodOptional<z.ZodString>;
             use_gpu: z.ZodOptional<z.ZodBoolean>;
@@ -294,7 +297,7 @@ export declare const requestSchema: z.ZodUnion<readonly [z.ZodObject<{
         lora?: string | undefined;
         predict?: number | undefined;
         no_mmap?: boolean | undefined;
-        verbosity?: 0 | 1 | -1 | undefined;
+        verbosity?: 0 | 2 | 1 | 3 | undefined;
         presence_penalty?: number | undefined;
         frequency_penalty?: number | undefined;
         repeat_penalty?: number | undefined;
@@ -363,7 +366,7 @@ export declare const requestSchema: z.ZodUnion<readonly [z.ZodObject<{
             speech_pad_ms?: number | undefined;
             samples_overlap?: number | undefined;
         } | undefined;
-        audio_format?: string | undefined;
+        audio_format?: "f32le" | "s16le" | undefined;
         contextParams?: {
             model?: string | undefined;
             use_gpu?: boolean | undefined;
@@ -469,7 +472,7 @@ export declare const requestSchema: z.ZodUnion<readonly [z.ZodObject<{
         lora?: string | undefined;
         predict?: number | undefined;
         no_mmap?: boolean | undefined;
-        verbosity?: 0 | 1 | -1 | undefined;
+        verbosity?: 0 | 2 | 1 | 3 | undefined;
         presence_penalty?: number | undefined;
         frequency_penalty?: number | undefined;
         repeat_penalty?: number | undefined;
@@ -538,7 +541,7 @@ export declare const requestSchema: z.ZodUnion<readonly [z.ZodObject<{
             speech_pad_ms?: number | undefined;
             samples_overlap?: number | undefined;
         } | undefined;
-        audio_format?: string | undefined;
+        audio_format?: "f32le" | "s16le" | undefined;
         contextParams?: {
             model?: string | undefined;
             use_gpu?: boolean | undefined;
@@ -629,7 +632,77 @@ export declare const requestSchema: z.ZodUnion<readonly [z.ZodObject<{
         fallbackToLocal: boolean;
         timeout?: number | undefined;
     } | undefined;
-}>>, z.ZodPipe<z.ZodObject<{
+}>>, z.ZodDiscriminatedUnion<[z.ZodObject<{
+    type: z.ZodLiteral<"loadModel">;
+    modelId: z.ZodString;
+    modelSrc: z.ZodOptional<z.ZodNever>;
+    withProgress: z.ZodOptional<z.ZodNever>;
+    delegate: z.ZodOptional<z.ZodNever>;
+    seed: z.ZodOptional<z.ZodNever>;
+    modelType: z.ZodLiteral<"whisper">;
+    modelConfig: z.ZodObject<{
+        strategy: z.ZodOptional<z.ZodOptional<z.ZodEnum<{
+            greedy: "greedy";
+            beam_search: "beam_search";
+        }>>>;
+        n_threads: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
+        n_max_text_ctx: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
+        offset_ms: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
+        duration_ms: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
+        audio_ctx: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
+        translate: z.ZodOptional<z.ZodOptional<z.ZodBoolean>>;
+        no_context: z.ZodOptional<z.ZodOptional<z.ZodBoolean>>;
+        no_timestamps: z.ZodOptional<z.ZodOptional<z.ZodBoolean>>;
+        single_segment: z.ZodOptional<z.ZodOptional<z.ZodBoolean>>;
+        print_special: z.ZodOptional<z.ZodOptional<z.ZodBoolean>>;
+        print_progress: z.ZodOptional<z.ZodOptional<z.ZodBoolean>>;
+        print_realtime: z.ZodOptional<z.ZodOptional<z.ZodBoolean>>;
+        print_timestamps: z.ZodOptional<z.ZodOptional<z.ZodBoolean>>;
+        token_timestamps: z.ZodOptional<z.ZodOptional<z.ZodBoolean>>;
+        thold_pt: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
+        thold_ptsum: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
+        max_len: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
+        split_on_word: z.ZodOptional<z.ZodOptional<z.ZodBoolean>>;
+        max_tokens: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
+        debug_mode: z.ZodOptional<z.ZodOptional<z.ZodBoolean>>;
+        tdrz_enable: z.ZodOptional<z.ZodOptional<z.ZodBoolean>>;
+        suppress_regex: z.ZodOptional<z.ZodOptional<z.ZodString>>;
+        initial_prompt: z.ZodOptional<z.ZodOptional<z.ZodString>>;
+        language: z.ZodOptional<z.ZodOptional<z.ZodString>>;
+        detect_language: z.ZodOptional<z.ZodOptional<z.ZodBoolean>>;
+        suppress_blank: z.ZodOptional<z.ZodOptional<z.ZodBoolean>>;
+        suppress_nst: z.ZodOptional<z.ZodOptional<z.ZodBoolean>>;
+        temperature: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
+        length_penalty: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
+        temperature_inc: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
+        entropy_thold: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
+        logprob_thold: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
+        greedy_best_of: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
+        beam_search_beam_size: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
+        vad_model_path: z.ZodOptional<z.ZodOptional<z.ZodString>>;
+        vad_params: z.ZodOptional<z.ZodOptional<z.ZodObject<{
+            threshold: z.ZodOptional<z.ZodNumber>;
+            min_speech_duration_ms: z.ZodOptional<z.ZodNumber>;
+            min_silence_duration_ms: z.ZodOptional<z.ZodNumber>;
+            max_speech_duration_s: z.ZodOptional<z.ZodNumber>;
+            speech_pad_ms: z.ZodOptional<z.ZodNumber>;
+            samples_overlap: z.ZodOptional<z.ZodNumber>;
+        }, z.core.$strip>>>;
+        audio_format: z.ZodOptional<z.ZodOptional<z.ZodEnum<{
+            f32le: "f32le";
+            s16le: "s16le";
+        }>>>;
+        contextParams: z.ZodOptional<z.ZodOptional<z.ZodObject<{
+            model: z.ZodOptional<z.ZodString>;
+            use_gpu: z.ZodOptional<z.ZodBoolean>;
+            flash_attn: z.ZodOptional<z.ZodBoolean>;
+            gpu_device: z.ZodOptional<z.ZodNumber>;
+        }, z.core.$strip>>>;
+        miscConfig: z.ZodOptional<z.ZodOptional<z.ZodObject<{
+            caption_enabled: z.ZodOptional<z.ZodBoolean>;
+        }, z.core.$strip>>>;
+    }, z.core.$strip>;
+}, z.core.$strip>], "modelType">]>, z.ZodPipe<z.ZodObject<{
     type: z.ZodLiteral<"downloadAsset">;
     assetSrc: z.ZodString;
     withProgress: z.ZodOptional<z.ZodBoolean>;
@@ -693,11 +766,11 @@ export declare const requestSchema: z.ZodUnion<readonly [z.ZodObject<{
     prompt: z.ZodOptional<z.ZodString>;
     type: z.ZodLiteral<"transcribeStream">;
 }, z.core.$strip>, z.ZodObject<{
-    modelId: z.ZodString;
+    id: z.ZodString;
     type: z.ZodLiteral<"loggingStream">;
 }, z.core.$strip>, z.ZodObject<{
     modelId: z.ZodString;
-    text: z.ZodString;
+    text: z.ZodUnion<readonly [z.ZodString, z.ZodArray<z.ZodString>]>;
     type: z.ZodLiteral<"embed">;
 }, z.core.$strip>, z.ZodUnion<readonly [z.ZodObject<{
     modelId: z.ZodString;
@@ -786,12 +859,6 @@ export declare const requestSchema: z.ZodUnion<readonly [z.ZodObject<{
     kvCacheKey: z.ZodString;
     modelId: z.ZodOptional<z.ZodString>;
 }, z.core.$strip>]>, z.ZodObject<{
-    type: z.ZodLiteral<"setConfig">;
-    config: z.ZodObject<{
-        cacheDirectory: z.ZodOptional<z.ZodString>;
-        swarmRelays: z.ZodOptional<z.ZodArray<z.ZodString>>;
-    }, z.core.$strip>;
-}, z.core.$strip>, z.ZodObject<{
     name: z.ZodString;
     type: z.ZodLiteral<"getModelInfo">;
 }, z.core.$strip>]>;
@@ -871,7 +938,7 @@ export declare const responseSchema: z.ZodDiscriminatedUnion<[z.ZodObject<{
     error: z.ZodOptional<z.ZodString>;
 }, z.core.$strip>, z.ZodObject<{
     type: z.ZodLiteral<"loggingStream">;
-    modelId: z.ZodString;
+    id: z.ZodString;
     level: z.ZodEnum<{
         error: "error";
         warn: "warn";
@@ -885,7 +952,7 @@ export declare const responseSchema: z.ZodDiscriminatedUnion<[z.ZodObject<{
 }, z.core.$strip>, z.ZodObject<{
     type: z.ZodLiteral<"embed">;
     success: z.ZodBoolean;
-    embedding: z.ZodDefault<z.ZodArray<z.ZodNumber>>;
+    embedding: z.ZodDefault<z.ZodUnion<readonly [z.ZodArray<z.ZodNumber>, z.ZodArray<z.ZodArray<z.ZodNumber>>]>>;
     error: z.ZodOptional<z.ZodString>;
 }, z.core.$strip>, z.ZodObject<{
     type: z.ZodLiteral<"translate">;
@@ -954,10 +1021,6 @@ export declare const responseSchema: z.ZodDiscriminatedUnion<[z.ZodObject<{
     type: z.ZodLiteral<"deleteCache">;
     success: z.ZodBoolean;
     error: z.ZodOptional<z.ZodString>;
-}, z.core.$strip>, z.ZodObject<{
-    type: z.ZodLiteral<"setConfig">;
-    success: z.ZodBoolean;
-    error: z.ZodOptional<z.ZodString>;
 }, z.core.$strip>, z.ZodObject<{
     type: z.ZodLiteral<"getModelInfo">;
     modelInfo: z.ZodObject<{
@@ -972,8 +1035,8 @@ export declare const responseSchema: z.ZodDiscriminatedUnion<[z.ZodObject<{
             whisper: "whisper";
             embeddings: "embeddings";
             nmt: "nmt";
-            vad: "vad";
             tts: "tts";
+            vad: "vad";
         }>;
         isCached: z.ZodBoolean;
         isLoaded: z.ZodBoolean;

package/dist/schemas/common.js

@@ -15,7 +15,6 @@ import { ttsRequestSchema, ttsResponseSchema } from "./text-to-speech.js";
 import { errorResponseSchema } from "./error.js";
 import { ragRequestSchema, ragResponseSchema } from "./rag.js";
 import { deleteCacheRequestSchema, deleteCacheResponseSchema, } from "./delete-cache.js";
-import { setConfigRequestSchema, setConfigResponseSchema } from "./set-config.js";
 import { getModelInfoRequestSchema, getModelInfoResponseSchema, } from "./get-model-info.js";
 export const requestSchema = z.union([
     pingRequestSchema,
@@ -33,7 +32,6 @@ export const requestSchema = z.union([
     stopProvideRequestSchema,
     ragRequestSchema,
     deleteCacheRequestSchema,
-    setConfigRequestSchema,
     getModelInfoRequestSchema,
 ]);
 export const responseSchema = z.discriminatedUnion("type", [
@@ -54,7 +52,6 @@ export const responseSchema = z.discriminatedUnion("type", [
     errorResponseSchema,
     ragResponseSchema,
     deleteCacheResponseSchema,
-    setConfigResponseSchema,
     getModelInfoResponseSchema,
 ]);
 export const rpcOptionsSchema = z.object({

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

@@ -14,8 +14,8 @@ export declare const downloadAssetOptionsSchema: z.ZodPipe<z.ZodObject<{
             whisper: "whisper";
             embeddings: "embeddings";
             nmt: "nmt";
-            vad: "vad";
             tts: "tts";
+            vad: "vad";
         }>>;
     }, z.core.$strip>]>;
     seed: z.ZodOptional<z.ZodBoolean>;
@@ -29,7 +29,7 @@ export declare const downloadAssetOptionsSchema: z.ZodPipe<z.ZodObject<{
         hyperbeeKey?: string | undefined;
         expectedSize?: number | undefined;
         sha256Checksum?: string | undefined;
-        addon?: "llm" | "whisper" | "embeddings" | "nmt" | "vad" | "tts" | undefined;
+        addon?: "llm" | "whisper" | "embeddings" | "nmt" | "tts" | "vad" | undefined;
     };
 }, {
     assetSrc: string | {
@@ -40,7 +40,7 @@ export declare const downloadAssetOptionsSchema: z.ZodPipe<z.ZodObject<{
         hyperbeeKey?: string | undefined;
         expectedSize?: number | undefined;
         sha256Checksum?: string | undefined;
-        addon?: "llm" | "whisper" | "embeddings" | "nmt" | "vad" | "tts" | undefined;
+        addon?: "llm" | "whisper" | "embeddings" | "nmt" | "tts" | "vad" | undefined;
     };
     seed?: boolean | undefined;
 }>>;
@@ -58,8 +58,8 @@ export declare const downloadAssetOptionsToRequestSchema: z.ZodPipe<z.ZodObject<
             whisper: "whisper";
             embeddings: "embeddings";
             nmt: "nmt";
-            vad: "vad";
             tts: "tts";
+            vad: "vad";
         }>>;
     }, z.core.$strip>]>;
     seed: z.ZodOptional<z.ZodBoolean>;
@@ -79,7 +79,7 @@ export declare const downloadAssetOptionsToRequestSchema: z.ZodPipe<z.ZodObject<
         hyperbeeKey?: string | undefined;
         expectedSize?: number | undefined;
         sha256Checksum?: string | undefined;
-        addon?: "llm" | "whisper" | "embeddings" | "nmt" | "vad" | "tts" | undefined;
+        addon?: "llm" | "whisper" | "embeddings" | "nmt" | "tts" | "vad" | undefined;
     };
     seed?: boolean | undefined;
     onProgress?: unknown;

package/dist/schemas/embed.d.ts

@@ -1,17 +1,17 @@
 import { z } from "zod";
 export declare const embedParamsSchema: z.ZodObject<{
     modelId: z.ZodString;
-    text: z.ZodString;
+    text: z.ZodUnion<readonly [z.ZodString, z.ZodArray<z.ZodString>]>;
 }, z.core.$strip>;
 export declare const embedRequestSchema: z.ZodObject<{
     modelId: z.ZodString;
-    text: z.ZodString;
+    text: z.ZodUnion<readonly [z.ZodString, z.ZodArray<z.ZodString>]>;
     type: z.ZodLiteral<"embed">;
 }, z.core.$strip>;
 export declare const embedResponseSchema: z.ZodObject<{
     type: z.ZodLiteral<"embed">;
     success: z.ZodBoolean;
-    embedding: z.ZodDefault<z.ZodArray<z.ZodNumber>>;
+    embedding: z.ZodDefault<z.ZodUnion<readonly [z.ZodArray<z.ZodNumber>, z.ZodArray<z.ZodArray<z.ZodNumber>>]>>;
     error: z.ZodOptional<z.ZodString>;
 }, z.core.$strip>;
 export type EmbedParams = z.infer<typeof embedParamsSchema>;

package/dist/schemas/embed.js

@@ -1,7 +1,12 @@
 import { z } from "zod";
 export const embedParamsSchema = z.object({
     modelId: z.string(),
-    text: z.string().min(1, "Text cannot be empty"),
+    text: z.union([
+        z.string().min(1, "Text cannot be empty"),
+        z
+            .array(z.string().min(1, "Text cannot be empty"))
+            .min(1, "Text array cannot be empty"),
+    ]),
 });
 export const embedRequestSchema = embedParamsSchema.extend({
     type: z.literal("embed"),
@@ -9,6 +14,8 @@ export const embedRequestSchema = embedParamsSchema.extend({
 export const embedResponseSchema = z.object({
     type: z.literal("embed"),
     success: z.boolean(),
-    embedding: z.array(z.number()).default([]),
+    embedding: z
+        .union([z.array(z.number()), z.array(z.array(z.number()))])
+        .default([]),
     error: z.string().optional(),
 });

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

@@ -32,8 +32,8 @@ export declare const modelInfoSchema: z.ZodObject<{
         whisper: "whisper";
         embeddings: "embeddings";
         nmt: "nmt";
-        vad: "vad";
         tts: "tts";
+        vad: "vad";
     }>;
     isCached: z.ZodBoolean;
     isLoaded: z.ZodBoolean;
@@ -68,8 +68,8 @@ export declare const getModelInfoResponseSchema: z.ZodObject<{
             whisper: "whisper";
             embeddings: "embeddings";
             nmt: "nmt";
-            vad: "vad";
             tts: "tts";
+            vad: "vad";
         }>;
         isCached: z.ZodBoolean;
         isLoaded: z.ZodBoolean;

package/dist/schemas/index.d.ts

@@ -6,6 +6,7 @@ export * from "./delete-cache";
 export * from "./download-asset";
 export * from "./embed";
 export * from "./load-model";
+export * from "./reload-config";
 export * from "./logging-stream";
 export * from "./provide";
 export * from "./stop-provide";
@@ -22,7 +23,10 @@ export * from "./error";
 export * from "./rag";
 export { SDK_CLIENT_ERROR_CODES } from "./sdk-errors-client";
 export { SDK_SERVER_ERROR_CODES } from "./sdk-errors-server";
-export * from "./set-config";
+export { qvacConfigSchema, type QvacConfig } from "./sdk-config";
 export * from "./get-model-info";
 export * from "./model-src-utils";
+export * from "./json-schema";
+export { type McpClient, type McpClientInput } from "./mcp-adapter";
+export { MODEL_TYPES, type ModelType } from "./model-src-utils";
 //# sourceMappingURL=index.d.ts.map

package/dist/schemas/index.js

@@ -7,6 +7,7 @@ export * from "./delete-cache.js";
 export * from "./download-asset.js";
 export * from "./embed.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";
@@ -23,6 +24,9 @@ export * from "./error.js";
 export * from "./rag.js";
 export { SDK_CLIENT_ERROR_CODES } from "./sdk-errors-client.js";
 export { SDK_SERVER_ERROR_CODES } from "./sdk-errors-server.js";
-export * from "./set-config.js";
+export { qvacConfigSchema } from "./sdk-config.js";
 export * from "./get-model-info.js";
 export * from "./model-src-utils.js";
+export * from "./json-schema.js";
+export {} from "./mcp-adapter.js";
+export { MODEL_TYPES } from "./model-src-utils.js";

package/dist/schemas/json-schema.d.ts

@@ -0,0 +1,16 @@
+export type JsonSchemaType = "string" | "number" | "integer" | "boolean" | "object" | "array" | "null";
+export type JsonSchemaProperty = {
+    type?: JsonSchemaType | JsonSchemaType[];
+    description?: string;
+    enum?: (string | number | boolean | null)[];
+    items?: JsonSchema;
+    properties?: Record<string, JsonSchemaProperty>;
+    required?: string[];
+    default?: unknown;
+    additionalProperties?: boolean | JsonSchemaProperty;
+};
+export type JsonSchema = JsonSchemaProperty & {
+    $schema?: string;
+    title?: string;
+};
+//# sourceMappingURL=json-schema.d.ts.map

package/dist/schemas/json-schema.js

@@ -0,0 +1,2 @@
+// ============== JSON Schema Types ==============
+export {};

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

@@ -1,8 +1,9 @@
 import { z } from "zod";
 export declare const VERBOSITY: {
-    readonly SILENT: -1;
-    readonly INFO: 0;
-    readonly DEBUG: 1;
+    readonly ERROR: 0;
+    readonly WARN: 1;
+    readonly INFO: 2;
+    readonly DEBUG: 3;
 };
 export declare const llmConfigSchema: z.ZodObject<{
     ctx_size: z.ZodDefault<z.ZodNumber>;
@@ -16,7 +17,7 @@ export declare const llmConfigSchema: z.ZodObject<{
     predict: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<-1>, z.ZodLiteral<-2>, z.ZodNumber]>>;
     system_prompt: z.ZodDefault<z.ZodString>;
     no_mmap: z.ZodOptional<z.ZodBoolean>;
-    verbosity: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<-1>, z.ZodLiteral<0>, z.ZodLiteral<1>]>>;
+    verbosity: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<0>, z.ZodLiteral<1>, z.ZodLiteral<2>, z.ZodLiteral<3>]>>;
     presence_penalty: z.ZodOptional<z.ZodNumber>;
     frequency_penalty: z.ZodOptional<z.ZodNumber>;
     repeat_penalty: z.ZodOptional<z.ZodNumber>;

package/dist/schemas/llamacpp-config.js

@@ -1,11 +1,13 @@
 import { z } from "zod";
 export const VERBOSITY = {
-    SILENT: -1,
-    INFO: 0,
-    DEBUG: 1,
+    ERROR: 0,
+    WARN: 1,
+    INFO: 2,
+    DEBUG: 3,
 };
 const verbositySchema = z.union([
-    z.literal(VERBOSITY.SILENT),
+    z.literal(VERBOSITY.ERROR),
+    z.literal(VERBOSITY.WARN),
     z.literal(VERBOSITY.INFO),
     z.literal(VERBOSITY.DEBUG),
 ]);
@@ -37,5 +39,5 @@ export const llmConfigSchema = z.object({
 });
 export const embedConfigSchema = z.object({
     // https://github.com/tetherto/qvac-lib-infer-llamacpp-embed/tree/main?tab=readme-ov-file#4-create-config
-    config: z.string().default("-ngl\t99\n-dev\tgpu"),
+    config: z.string().default("-ngl\t99\n-dev\tgpu\n--batch_size\t1024"),
 });