@qvac/sdk 0.16.0 → 0.17.0

package/NOTICE

@@ -61,24 +61,24 @@ Third-Party Model Licenses
 
 --- apache-2.0 (Apache License 2.0) ---
 
+  abot-world-0-5b-lf-dit-q8_0
+    https://huggingface.co/acvlab/ABot-World-0-5B-LF
   bci-embedder
     https://github.com/tetherto/qvac/releases/download/bci-test-assets-v0.1.0/bci-embedder.bin
+  cosyvoice3-flow-f32
+    https://huggingface.co/FunAudioLLM/Fun-CosyVoice3-0.5B-2512
+  cosyvoice3-hift-f32
+    https://huggingface.co/FunAudioLLM/Fun-CosyVoice3-0.5B-2512
+  cosyvoice3-llm-q8_0
+    https://huggingface.co/FunAudioLLM/Fun-CosyVoice3-0.5B-2512
   craft_mlt_25k
     https://www.jaided.ai/easyocr/modelhub/
   crnn_mobilenet_v3_small
     https://github.com/mindee/doctr
-  crnn_mobilenet_v3_small
-    https://github.com/felixdittrich92/OnnxTR
   db_mobilenet_v3_large
     https://github.com/mindee/doctr
-  db_mobilenet_v3_large
-    https://github.com/felixdittrich92/OnnxTR
-  db_resnet50
-    https://github.com/felixdittrich92/OnnxTR
   de-tiny-ggml-model-f16
     https://huggingface.co/primeline/whisper-tiny-german-1224
-  detector_craft
-    https://www.jaided.ai/easyocr/modelhub/
   dolphin-mixtral-2x7b-dop-Q2_K-00001-of-00005
     https://huggingface.co/jmb95/laser-dolphin-mixtral-2x7b-dpo-GGUF
   en-base-ggml-model-f16
@@ -113,6 +113,8 @@ Third-Party Model Licenses
     https://huggingface.co/bygreencn/whisper-base-ja
   ja-tiny-ggml-model-f16
     https://huggingface.co/CheeLi03/whisper-tiny-ja-puct-4k
+  Kwaipilot_KAT-Coder-V2.5-Dev-GGUF
+    https://huggingface.co/bartowski/Kwaipilot_KAT-Coder-V2.5-Dev-GGUF
   laser-dolphin-mixtral-2x7b-dpo-GGUF
     https://huggingface.co/jmb95/laser-dolphin-mixtral-2x7b-dpo-GGUF
   latin_g2
@@ -125,6 +127,10 @@ Third-Party Model Licenses
     https://huggingface.co/qvac/MedPsy-1.7B-GGUF
   MedPsy-4B-GGUF
     https://huggingface.co/qvac/MedPsy-4B-GGUF
+  merges
+    https://huggingface.co/FunAudioLLM/Fun-CosyVoice3-0.5B-2512
+  mmproj-unlimited-ocr-Q8_0
+    https://huggingface.co/baidu/Unlimited-OCR
   nb-tiny-ggml-model
     https://huggingface.co/NbAiLab/nb-whisper-tiny
   parler-indic-f16
@@ -133,8 +139,6 @@ Third-Party Model Licenses
     https://huggingface.co/parler-tts/parler-tts-large-v1
   parler-mini-v1-f16
     https://huggingface.co/parler-tts/parler-tts-mini-v1
-  parseq
-    https://github.com/felixdittrich92/OnnxTR
   pi05-base-q-aggressive
     https://huggingface.co/lerobot/pi05_base
   pt-base-ggml-model-f16
@@ -175,32 +179,6 @@ Third-Party Model Licenses
     https://huggingface.co/unsloth/Qwen3.6-27B-GGUF
   Qwen3.6-35B-A3B-GGUF
     https://huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF
-  recognizer_arabic
-    https://www.jaided.ai/easyocr/modelhub/
-  recognizer_bengali
-    https://www.jaided.ai/easyocr/modelhub/
-  recognizer_cyrillic
-    https://www.jaided.ai/easyocr/modelhub/
-  recognizer_devanagari
-    https://www.jaided.ai/easyocr/modelhub/
-  recognizer_japanese
-    https://www.jaided.ai/easyocr/modelhub/
-  recognizer_kannada
-    https://www.jaided.ai/easyocr/modelhub/
-  recognizer_korean
-    https://www.jaided.ai/easyocr/modelhub/
-  recognizer_latin
-    https://www.jaided.ai/easyocr/modelhub/
-  recognizer_tamil
-    https://www.jaided.ai/easyocr/modelhub/
-  recognizer_telugu
-    https://www.jaided.ai/easyocr/modelhub/
-  recognizer_thai
-    https://www.jaided.ai/easyocr/modelhub/
-  recognizer_zh_sim
-    https://www.jaided.ai/easyocr/modelhub/
-  recognizer_zh_tra
-    https://www.jaided.ai/easyocr/modelhub/
   ru-base-ggml-model-f16
     https://huggingface.co/CheeLi03/whisper-base-rus-8
   ru-tiny-ggml-model-f16
@@ -217,14 +195,30 @@ Third-Party Model Licenses
     https://huggingface.co/HuggingFaceVLA/smolvla_libero
   SmolVLM2-500M-Video-Instruct-GGUF
     https://huggingface.co/ggml-org/SmolVLM2-500M-Video-Instruct-GGUF
+  umt5-xxl-enc-q8_0
+    https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged
+  unlimited-ocr-gguf
+    https://huggingface.co/vimalnakrani/unlimited-ocr-gguf
+  voice-en
+    https://huggingface.co/FunAudioLLM/Fun-CosyVoice3-0.5B-2512
+  voice-zh-2
+    https://huggingface.co/FunAudioLLM/Fun-CosyVoice3-0.5B-2512
+  voice-zh-female
+    https://huggingface.co/FunAudioLLM/Fun-CosyVoice3-0.5B-2512
   Wan_2.1_ComfyUI_repackaged
     https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged
+  Wan_2.2_ComfyUI_Repackaged
+    https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged
   Wan2.1-I2V-14B-480P-gguf
     https://huggingface.co/city96/Wan2.1-I2V-14B-480P-gguf
   Wan2.1-I2V-14B-720P-gguf
     https://huggingface.co/city96/Wan2.1-I2V-14B-720P-gguf
   Wan2.1-T2V-14B-gguf
     https://huggingface.co/city96/Wan2.1-T2V-14B-gguf
+  wan2.2_vae_f16
+    https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged
+  Wan2.2-TI2V-5B-Turbo-GGUF
+    https://huggingface.co/hum-ma/Wan2.2-TI2V-5B-Turbo-GGUF
 
 --- bsd-3-clause (BSD 3-Clause License) ---
 
@@ -280,6 +274,8 @@ Third-Party Model Licenses
 
   embeddinggemma-300m-GGUF
     https://huggingface.co/unsloth/embeddinggemma-300m-GGUF
+  gemma-3-12b-it-GGUF
+    https://huggingface.co/unsloth/gemma-3-12b-it-GGUF
 
 --- health-ai-developer-foundations (HAI-DEF Terms of Use) ---
 
@@ -297,6 +293,13 @@ Third-Party Model Licenses
   Llama-3.2-1B-Instruct-Q4_0-00001-of-00008
     https://huggingface.co/bartowski/Llama-3.2-1B-Instruct-GGUF
 
+--- ltx-2-community ---
+
+  LTX-2.3-GGUF
+    https://huggingface.co/QuantStack/LTX-2.3-GGUF
+  LTX-2.3-GGUF
+    https://huggingface.co/unsloth/LTX-2.3-GGUF
+
 --- mit (MIT License) ---
 
   acestep-5Hz-lm-0.6B-Q8_0
@@ -325,6 +328,8 @@ Third-Party Model Licenses
     https://huggingface.co/ResembleAI/chatterbox-turbo
   de-base-ggml-model-f16
     https://huggingface.co/aware-ai/whisper-base-german
+  DeepSeek-V4-Flash-0731-GGUF
+    https://huggingface.co/unsloth/DeepSeek-V4-Flash-0731-GGUF
   es-tiny-ggml-model-f16
     https://huggingface.co/JaepaX/whisper-tiny-es
   ggml-indictrans2-en-indic-1B-f16
@@ -349,6 +354,8 @@ Third-Party Model Licenses
     https://huggingface.co/spaces/YatharthS
   pt-tiny-ggml-model-f16
     https://huggingface.co/DavyCosta701/whisper-tiny-pt-bpra
+  taew2_2_f16
+    https://github.com/madebyollin/taehv
   tiny_acft_q8_0
     https://voiceinput.futo.org/VoiceInput/tiny_en_acft_q8_0.bin
   tiny_en_acft_q8_0
@@ -577,6 +584,8 @@ Third-Party Model Licenses
     https://huggingface.co/nvidia/diar_streaming_sortformer_4spk-v2.1
   groot-q8_vf16
     https://huggingface.co/nvidia/GR00T-N1.7-LIBERO
+  groot-q8_vf16
+    https://huggingface.co/nvidia/GR00T-N1.7-3B
 
 --- openrail ---
 
@@ -623,29 +632,37 @@ JavaScript Dependencies
 
   @hyperswarm/secret-stream@6.9.1
     https://github.com/holepunchto/hyperswarm-secret-stream
-  @qvac/bci-whispercpp@0.5.1
+  @qvac/asr-ggml@0.1.1
     https://github.com/tetherto/qvac
-  @qvac/classification-ggml@0.13.0
+  @qvac/audiogen-ggml@0.1.1
+    https://github.com/tetherto/qvac
+  @qvac/bci-whispercpp@0.6.0
+    https://github.com/tetherto/qvac
+  @qvac/classification-ggml@0.15.1
     https://github.com/tetherto/qvac
   @qvac/decoder-audio@0.5.0
     https://github.com/tetherto/qvac
   @qvac/diagnostics@0.1.2
     https://github.com/tetherto/qvac
-  @qvac/diffusion-cpp@0.15.0
+  @qvac/diffusion-cpp@0.17.0
     https://github.com/tetherto/qvac
-  @qvac/embed-llamacpp@0.28.0
+  @qvac/embed-llamacpp@0.30.1
     https://github.com/tetherto/qvac
   @qvac/error@0.1.1
+  @qvac/fabric@0.3.1
+    https://github.com/tetherto/qvac
   @qvac/infer-base@0.4.2
     https://github.com/tetherto/qvac
   @qvac/infer-base@0.6.1
     https://github.com/tetherto/qvac
+  @qvac/infer-base@0.6.2
+    https://github.com/tetherto/qvac
   @qvac/langdetect-text@0.1.2
     https://github.com/tetherto/qvac
-  @qvac/llm-llamacpp@0.38.0
+  @qvac/llm-llamacpp@0.39.4
     https://github.com/tetherto/qvac
   @qvac/logging@0.1.0
-  @qvac/ocr-ggml@0.12.1
+  @qvac/ocr-ggml@0.13.1
     https://github.com/tetherto/qvac
   @qvac/rag@0.6.4
     https://github.com/tetherto/qvac
@@ -653,15 +670,11 @@ JavaScript Dependencies
     https://github.com/tetherto/qvac
   @qvac/registry-schema@0.3.0
     https://github.com/tetherto/qvac
-  @qvac/transcription-parakeet@0.10.1
-    https://github.com/tetherto/qvac
-  @qvac/transcription-whispercpp@0.12.1
-    https://github.com/tetherto/qvac
-  @qvac/translation-nmtcpp@8.1.0
+  @qvac/translation-nmtcpp@8.3.1
     https://github.com/tetherto/qvac
-  @qvac/tts-ggml@0.5.1
+  @qvac/tts-ggml@0.6.3
     https://github.com/tetherto/qvac
-  @qvac/vla-ggml@0.14.0
+  @qvac/vla-ggml@0.16.2
     https://github.com/tetherto/qvac
   adaptive-timeout@1.0.1
     https://github.com/holepunchto/adaptive-timeout
@@ -683,6 +696,8 @@ JavaScript Dependencies
     https://github.com/holepunchto/bare-bundle
   bare-bundle-id@1.0.2
     https://github.com/holepunchto/bare-bundle-id
+  bare-cpu-info@0.1.1
+    https://github.com/holepunchto/bare-cpu-info
   bare-crypto@1.15.3
     https://github.com/holepunchto/bare-crypto
   bare-dns@2.1.4
@@ -693,12 +708,14 @@ JavaScript Dependencies
     https://github.com/holepunchto/bare-events
   bare-fetch@3.0.1
     https://github.com/holepunchto/bare-fetch
-  bare-ffmpeg@1.2.3
+  bare-ffmpeg@1.5.0
     https://github.com/holepunchto/bare-ffmpeg
   bare-form-data@1.2.2
     https://github.com/holepunchto/bare-form-data
   bare-fs@4.7.2
     https://github.com/holepunchto/bare-fs
+  bare-gpu-info@0.1.1
+    https://github.com/holepunchto/bare-gpu-info
   bare-hrtime@2.1.1
     https://github.com/holepunchto/bare-hrtime
   bare-http-parser@1.1.5

package/README.md

@@ -75,6 +75,71 @@ try {
 node quickstart.js
 ```
 
+## System resource diagnostics
+
+Use `getSystemResources` to inspect locally observed CPU, system-memory, GPU, and
+driver capabilities. Pass `sample: true` only when you also need a fresh usage
+sample:
+
+```ts
+import { getSystemResources } from '@qvac/sdk'
+
+const resources = await getSystemResources({ sample: true })
+
+if (resources.capabilities.memory.totalBytes.status === 'supported') {
+  console.log('System memory:', resources.capabilities.memory.totalBytes.value)
+}
+
+if (resources.sample?.cpu.status === 'supported') {
+  console.log('CPU utilization:', resources.sample.cpu.value)
+}
+```
+
+See the [system resources support matrix](./docs/system-resources-support-matrix.md)
+for metric-level evidence and platform limitations.
+
+Every metric reports `supported`, `unavailable`, `unverified`, or `failed`.
+Supported values include provenance with a source and optional scope. These
+values are diagnostics; they do not reserve memory or guarantee that a model
+can be loaded.
+
+GPU capabilities expose observed driver names, versions, and graphics APIs.
+These observations do not prove that an inference backend is compatible.
+
+Profiled inference operation events may include `event.backend` with the
+selected backend and device, graphics API, driver, fallback reason, and probe
+result. Addons attach backend metadata with `attachBackendDiagnostics`; the SDK
+validates it before recording the operation event. `gpuId`, when present,
+identifies a GPU from the current worker resource collector and is stable only
+for that collector's lifetime. The SDK does not infer compatibility from driver
+inventory or log text.
+
+### Profiler resource gauges
+
+Resource gauges are disabled by default. Enable them explicitly to attach one
+worker resource sample to each profiled operation:
+
+```ts
+import { profiler } from '@qvac/sdk'
+
+profiler.enable({ mode: 'verbose', includeResourceGauges: true })
+
+// Run SDK operations, then inspect recentEvents[].resources.
+const profile = profiler.exportJSON()
+```
+
+The sample uses the same status, provenance, and scope semantics as
+`getSystemResources({ sample: true })`. Its `sampledAt` uses the same monotonic
+clock as the profiling event's `ts`, so the two timestamps are comparable.
+`resources.origin` is `local` for the current worker and `provider` for a
+delegated provider's worker. Samples are delivered to `profiler.onRecord`; they
+are retained in `exportJSON().recentEvents` only in `verbose` mode. Enabling
+gauges in `summary` mode still incurs the sampling cost without retaining them.
+Disabling profiling or omitting `includeResourceGauges` performs no resource
+sampling. Enabling gauges adds one CPU query and one query per GPU to each
+profiled operation's response path. If the worker resource collector is not
+initialized, the event omits the resource block.
+
 ## Examples
 
 In the `./examples` subdirectory, you will find scripts demonstrating how to use all SDK functionalities. To try any of them:

package/dist/client/api/audio-gen-result.d.ts

@@ -0,0 +1,4 @@
+import { type AudioGenClientParams, type AudioGenResult, type AudioGenStreamRequest } from '../../schemas/audio-gen.js';
+export type AudioGenStreamFactory = (request: AudioGenStreamRequest) => AsyncGenerator<unknown>;
+export declare function createAudioGenResult(params: AudioGenClientParams, streamFactory: AudioGenStreamFactory): AudioGenResult;
+//# sourceMappingURL=audio-gen-result.d.ts.map

package/dist/client/api/audio-gen-result.js

@@ -0,0 +1,132 @@
+import { audioGenClientParamsSchema, audioGenStreamResponseSchema } from '../../schemas/audio-gen.js';
+import { parseClientInput } from '../../client/parse-input.js';
+import { generateClientRequestId } from '../../client/api/client-request-id.js';
+import { decodeBase64 } from '../../utils/encoding.js';
+import { InvalidResponseError } from '../../utils/errors-client.js';
+import { InferenceCancelledError } from '../../utils/errors-server.js';
+function concatenateChunks(chunks) {
+    const totalLength = chunks.reduce((total, chunk) => total + chunk.length, 0);
+    const pcm = new Uint8Array(totalLength);
+    let offset = 0;
+    for (const chunk of chunks) {
+        pcm.set(chunk, offset);
+        offset += chunk.length;
+    }
+    return pcm;
+}
+export function createAudioGenResult(params, streamFactory) {
+    const parsed = parseClientInput(audioGenClientParamsSchema, params);
+    const requestId = generateClientRequestId();
+    const request = {
+        ...parsed,
+        type: 'audioGenStream',
+        requestId
+    };
+    const progressQueue = [];
+    const pcmChunks = [];
+    let sampleRate;
+    let channels;
+    let bitsPerSample;
+    let progressDone = false;
+    let progressError;
+    let progressResolve;
+    let resolveAudio = () => { };
+    let rejectAudio = () => { };
+    const audio = new Promise((resolve, reject) => {
+        resolveAudio = resolve;
+        rejectAudio = reject;
+    });
+    audio.catch(() => { });
+    let resolveStats = () => { };
+    let rejectStats = () => { };
+    const stats = new Promise((resolve, reject) => {
+        resolveStats = resolve;
+        rejectStats = reject;
+    });
+    stats.catch(() => { });
+    function notifyProgress() {
+        progressResolve?.();
+        progressResolve = undefined;
+    }
+    async function processResponses() {
+        let receivedDone = false;
+        try {
+            for await (const response of streamFactory(request)) {
+                if (!response ||
+                    typeof response !== 'object' ||
+                    !('type' in response) ||
+                    response.type !== 'audioGenStream') {
+                    continue;
+                }
+                const chunk = audioGenStreamResponseSchema.parse(response);
+                if (chunk.progress) {
+                    progressQueue.push(chunk.progress);
+                    notifyProgress();
+                }
+                if (chunk.data !== undefined) {
+                    pcmChunks.push(decodeBase64(chunk.data));
+                    sampleRate = chunk.sampleRate;
+                    channels = chunk.channels;
+                    bitsPerSample = chunk.bitsPerSample;
+                }
+                if (chunk.done) {
+                    receivedDone = true;
+                    if (chunk.stopReason === 'cancelled') {
+                        const error = new InferenceCancelledError(requestId);
+                        rejectAudio(error);
+                        rejectStats(error);
+                        break;
+                    }
+                    if (sampleRate === undefined || channels === undefined || bitsPerSample === undefined) {
+                        throw new InvalidResponseError('audioGenStream audio chunk');
+                    }
+                    resolveAudio({
+                        pcm: concatenateChunks(pcmChunks),
+                        sampleRate,
+                        channels,
+                        bitsPerSample
+                    });
+                    resolveStats(chunk.stats);
+                    break;
+                }
+            }
+            if (!receivedDone) {
+                throw new InvalidResponseError('audioGenStream terminal response');
+            }
+        }
+        catch (error) {
+            progressError =
+                error instanceof Error ? error : new InvalidResponseError('audioGenStream', error);
+            rejectAudio(progressError);
+            rejectStats(progressError);
+        }
+        finally {
+            progressDone = true;
+            notifyProgress();
+        }
+    }
+    async function* progressStream() {
+        while (true) {
+            const tick = progressQueue.shift();
+            if (tick) {
+                yield tick;
+                continue;
+            }
+            if (progressDone) {
+                if (progressError !== undefined)
+                    throw progressError;
+                return;
+            }
+            await new Promise((resolve) => {
+                progressResolve = resolve;
+            });
+        }
+    }
+    void processResponses();
+    return {
+        requestId,
+        progressStream: progressStream(),
+        audio,
+        stats
+    };
+}

package/dist/client/api/audio-gen.d.ts

@@ -0,0 +1,29 @@
+import type { AudioGenClientParams, AudioGenResult } from '../../schemas/audio-gen.js';
+/**
+ * Generates PCM audio using a loaded AudioGen model.
+ *
+ * @param params - Loaded model ID, required caption, and optional lyrics, musical controls, duration, and seed.
+ * @returns A run with a synchronous request ID, generation progress stream, PCM audio promise, and stats promise.
+ *
+ * @example
+ * ```typescript
+ * const modelId = await loadModel({
+ *   modelType: "audiogen",
+ *   modelConfig: {
+ *     textEncModelSrc,
+ *     lmModelSrc,
+ *     ditModelSrc,
+ *     vaeModelSrc,
+ *   },
+ * });
+ * const run = audioGen({ modelId, caption: "ambient electronic music", duration: 10 });
+ * stopButton.onclick = () => cancel({ requestId: run.requestId });
+ * for await (const progress of run.progressStream) {
+ *   console.log(progress.stage, progress.step, progress.total);
+ * }
+ * const { pcm, sampleRate, channels, bitsPerSample } = await run.audio;
+ * const stats = await run.stats;
+ * ```
+ */
+export declare function audioGen(params: AudioGenClientParams): AudioGenResult;
+//# sourceMappingURL=audio-gen.d.ts.map

package/dist/client/api/audio-gen.js

@@ -0,0 +1,31 @@
+import { stream as streamRpc } from '../../client/rpc/rpc-client.js';
+import { createAudioGenResult } from '../../client/api/audio-gen-result.js';
+/**
+ * Generates PCM audio using a loaded AudioGen model.
+ *
+ * @param params - Loaded model ID, required caption, and optional lyrics, musical controls, duration, and seed.
+ * @returns A run with a synchronous request ID, generation progress stream, PCM audio promise, and stats promise.
+ *
+ * @example
+ * ```typescript
+ * const modelId = await loadModel({
+ *   modelType: "audiogen",
+ *   modelConfig: {
+ *     textEncModelSrc,
+ *     lmModelSrc,
+ *     ditModelSrc,
+ *     vaeModelSrc,
+ *   },
+ * });
+ * const run = audioGen({ modelId, caption: "ambient electronic music", duration: 10 });
+ * stopButton.onclick = () => cancel({ requestId: run.requestId });
+ * for await (const progress of run.progressStream) {
+ *   console.log(progress.stage, progress.step, progress.total);
+ * }
+ * const { pcm, sampleRate, channels, bitsPerSample } = await run.audio;
+ * const stats = await run.stats;
+ * ```
+ */
+export function audioGen(params) {
+    return createAudioGenResult(params, streamRpc);
+}

package/dist/client/api/batch-completion.d.ts

@@ -1,5 +1,5 @@
-import { type BatchCompletionClientParams, type BatchCompletionRun, type BatchCompletionStreamRequest, type McpClientInput, type RPCOptions, type Tool } from '../../schemas/index';
-import { type ToolInput } from '../../utils/tool-helpers';
+import { type BatchCompletionClientParams, type BatchCompletionRun, type BatchCompletionStreamRequest, type McpClientInput, type RPCOptions, type Tool } from '../../schemas/index.js';
+import { type ToolInput } from '../../utils/tool-helpers.js';
 type BatchPromptParams = Omit<BatchCompletionClientParams['prompts'][number], 'tools'> & {
     tools?: Tool[] | ToolInput[];
     mcp?: McpClientInput[];

package/dist/client/api/batch-completion.js

@@ -1,5 +1,5 @@
-import { stream as streamRpc } from '../rpc/rpc-client.js';
-import { generateClientRequestId } from './client-request-id.js';
+import { stream as streamRpc } from '../../client/rpc/rpc-client.js';
+import { generateClientRequestId } from '../../client/api/client-request-id.js';
 import { getClientLogger } from '../../logging/index.js';
 import { batchCompletionStreamResponseSchema } from '../../schemas/index.js';
 import { buildFinalFromEvents } from '../../utils/aggregate-events.js';

package/dist/client/api/bci-transcribe.d.ts

@@ -1,4 +1,4 @@
-import { type BciTranscribeClientParams, type BciTranscribeStreamClientParams, type BciTranscribeStreamSession, type BciTranscribeStreamMetadataSession, type RPCOptions, type TranscribeSegment } from '../../schemas/index';
+import { type BciTranscribeClientParams, type BciTranscribeStreamClientParams, type BciTranscribeStreamSession, type BciTranscribeStreamMetadataSession, type RPCOptions, type TranscribeSegment } from '../../schemas/index.js';
 /**
  * Transcribe a neural-signal buffer with a loaded BCI model and return the
  * complete text. Accepts either a `.bin` file path or a raw neural buffer.

package/dist/client/api/bci-transcribe.js

@@ -1,10 +1,10 @@
 import { bciTranscribeResponseSchema, bciTranscribeClientParamsSchema, bciTranscribeStreamResponseSchema } from '../../schemas/index.js';
-import { stream, duplex } from '../rpc/rpc-client.js';
+import { stream, duplex } from '../../client/rpc/rpc-client.js';
 import { getClientLogger } from '../../logging/index.js';
 import { TranscriptionFailedError } from '../../utils/errors-client.js';
 import { decoratePromise } from '../../utils/decorate-promise.js';
-import { parseClientInput } from '../parse-input.js';
-import { generateClientRequestId } from './client-request-id.js';
+import { parseClientInput } from '../../client/parse-input.js';
+import { generateClientRequestId } from '../../client/api/client-request-id.js';
 const logger = getClientLogger();
 function buildBciTranscribeRequest(params, requestId) {
     return {

package/dist/client/api/cancel.d.ts

@@ -1,4 +1,4 @@
-import { type CancelClientInput } from '../../schemas/index';
+import { type CancelClientInput } from '../../schemas/index.js';
 /**
  * Cancels an ongoing operation.
  *

package/dist/client/api/cancel.js

@@ -1,4 +1,4 @@
-import { send } from '../rpc/rpc-client.js';
+import { send } from '../../client/rpc/rpc-client.js';
 import {} from '../../schemas/index.js';
 import { InvalidResponseError, CancelFailedError } from '../../utils/errors-client.js';
 /**

package/dist/client/api/classify.d.ts

@@ -1,4 +1,4 @@
-import { type ClassifyClientParams, type ClassificationResult } from '../../schemas/index';
+import { type ClassifyClientParams, type ClassificationResult } from '../../schemas/index.js';
 /**
  * Classifies an image using a loaded classification model.
  *

package/dist/client/api/classify.js

@@ -1,5 +1,5 @@
 import { classifyResponseSchema } from '../../schemas/index.js';
-import { stream as streamRpc } from '../rpc/rpc-client.js';
+import { stream as streamRpc } from '../../client/rpc/rpc-client.js';
 import { encodeBase64 } from '../../utils/encoding.js';
 /**
  * Classifies an image using a loaded classification model.

package/dist/client/api/completion-stream.d.ts

@@ -1,5 +1,5 @@
-import { type CompletionClientParams, type CompletionRun, type McpClientInput, type Tool, type RPCOptions } from '../../schemas/index';
-import { type ToolInput } from '../../utils/tool-helpers';
+import { type CompletionClientParams, type CompletionRun, type McpClientInput, type Tool, type RPCOptions } from '../../schemas/index.js';
+import { type ToolInput } from '../../utils/tool-helpers.js';
 type CompletionParams = Omit<CompletionClientParams, 'tools'> & {
     tools?: Tool[] | ToolInput[];
     mcp?: McpClientInput[];
@@ -28,7 +28,7 @@ type CompletionParams = Omit<CompletionClientParams, 'tools'> & {
  * @param params.mcp - Optional array of MCP client inputs for tool integration
  * @param params.captureThinking - Best-effort parsing of `<think>` blocks into `thinkingDelta` events; `final.raw.fullText` always preserves the original output
  * @param params.emitRawDeltas - When true, every raw model token is also emitted as a `rawDelta` event
- * @param params.toolDialect - Override the SDK's name-based dialect detection. Supported values: `"hermes"`, `"pythonic"`, `"json"`, `"harmony"`, `"qwen35"` (Qwen3.5/3.6), `"gemma4"`. Use when the auto-router doesn't recognise your model name. Drives both streaming frame detection and finalization parsing.
+ * @param params.toolDialect - Override the SDK's name-based dialect detection. Supported values: `"hermes"`, `"pythonic"`, `"json"`, `"harmony"`, `"qwen35"` (Qwen3.5/3.6), `"gemma4"`, `"dsml"` (DeepSeek V3.2/V4). Use when the auto-router doesn't recognise your model name. Drives both streaming frame detection and finalization parsing.
  * Common override case: Llama 3.x tool-calling fine-tunes that emit the native pythonic header (`<|start_header_id|>tool_call<|end_header_id|>...<|eot_id|>`).
  * @param params.responseFormat - Optional structured-output constraint applied to the model's output:
  *   - `{ type: "text" }` — no constraint (default behavior)

package/dist/client/api/completion-stream.js

@@ -1,11 +1,11 @@
-import { stream as streamRpc } from '../rpc/rpc-client.js';
+import { stream as streamRpc } from '../../client/rpc/rpc-client.js';
 import { getClientLogger } from '../../logging/index.js';
 import { completionStreamResponseSchema } from '../../schemas/index.js';
 import { CompletionFailedError, InferenceCancelledError } from '../../utils/errors-server.js';
 import { getMcpToolsWithHandlers } from '../../utils/mcp-adapter.js';
 import { validateTools } from '../../utils/tool-helpers.js';
 import { buildFinalFromEvents } from '../../utils/aggregate-events.js';
-import { generateClientRequestId } from './client-request-id.js';
+import { generateClientRequestId } from '../../client/api/client-request-id.js';
 const logger = getClientLogger();
 /**
  * Generates completion from a language model based on conversation history.
@@ -28,7 +28,7 @@ const logger = getClientLogger();
  * @param params.mcp - Optional array of MCP client inputs for tool integration
  * @param params.captureThinking - Best-effort parsing of `<think>` blocks into `thinkingDelta` events; `final.raw.fullText` always preserves the original output
  * @param params.emitRawDeltas - When true, every raw model token is also emitted as a `rawDelta` event
- * @param params.toolDialect - Override the SDK's name-based dialect detection. Supported values: `"hermes"`, `"pythonic"`, `"json"`, `"harmony"`, `"qwen35"` (Qwen3.5/3.6), `"gemma4"`. Use when the auto-router doesn't recognise your model name. Drives both streaming frame detection and finalization parsing.
+ * @param params.toolDialect - Override the SDK's name-based dialect detection. Supported values: `"hermes"`, `"pythonic"`, `"json"`, `"harmony"`, `"qwen35"` (Qwen3.5/3.6), `"gemma4"`, `"dsml"` (DeepSeek V3.2/V4). Use when the auto-router doesn't recognise your model name. Drives both streaming frame detection and finalization parsing.
  * Common override case: Llama 3.x tool-calling fine-tunes that emit the native pythonic header (`<|start_header_id|>tool_call<|end_header_id|>...<|eot_id|>`).
  * @param params.responseFormat - Optional structured-output constraint applied to the model's output:
  *   - `{ type: "text" }` — no constraint (default behavior)

package/dist/client/api/delete-cache.js

@@ -1,4 +1,4 @@
-import { send } from '../rpc/rpc-client.js';
+import { send } from '../../client/rpc/rpc-client.js';
 import { InvalidDeleteCacheParamsError, DeleteCacheFailedError } from '../../utils/errors-client.js';
 /**
  * Deletes KV cache files.

package/dist/client/api/diffusion.d.ts

@@ -1,4 +1,4 @@
-import { type DiffusionClientParams, type DiffusionStats } from '../../schemas/index';
+import { type DiffusionClientParams, type DiffusionStats } from '../../schemas/index.js';
 export interface DiffusionProgressTick {
     step: number;
     totalSteps: number;