@qvac/harness 0.7.1 → 0.7.2

package/CHANGELOG.md

@@ -8,6 +8,28 @@ While the major version is `0`, a breaking change bumps the minor version.
 
 ## [Unreleased]
 
+## [0.7.2] - 2026-10-01
+
+### Added
+
+- `web_fetch` attaches a plain-text file (.txt, .csv, .json…) to the chat
+  besides returning its text, so the user can open or save it and `exec` can
+  stage it to read past the returned text. The result carries its
+  `attachmentId`, `fileName` and `byteLength`. A body cut at the fetch cap is
+  not attached, and a host that cannot store files gets the text alone, as
+  with page images. (#147, #148)
+
+### Changed
+
+- Speech sounds cleaner on desktops. Devices on the `mid` profile and up now
+  speak with Supertonic 3 FP16 (`Supertonic3-FP16`, 207 MB) instead of Q8_0
+  (127 MB), at 16 inference steps instead of 5, which removes the artifacts
+  the 5-step output had. Synthesis takes about 2.7× as long, still around 7×
+  faster than real time on an M-series Mac. Phones (`low`) and `tiny` devices
+  keep Q4_0 at 5 steps. A host that prefetches `infraModelsFor` downloads the
+  FP16 model once; `Supertonic3-Q8_0` stays in the catalog, and now also runs
+  at 16 steps when a host asks for it by name. (#149)
+
 ## [0.7.1] - 2026-09-30
 
 ### Added
@@ -1019,7 +1041,8 @@ Tagged but never published to npm; these changes first shipped in 0.2.12.
 - GPU stats are read through `bare-gpu-info`, so every dependency installs from
   the public npm registry. (#12)
 
-[Unreleased]: https://github.com/tetherto/qvac-harness/compare/v0.7.1...HEAD
+[Unreleased]: https://github.com/tetherto/qvac-harness/compare/v0.7.2...HEAD
+[0.7.2]: https://github.com/tetherto/qvac-harness/compare/v0.7.1...v0.7.2
 [0.7.1]: https://github.com/tetherto/qvac-harness/compare/v0.7.0...v0.7.1
 [0.7.0]: https://github.com/tetherto/qvac-harness/compare/v0.6.0...v0.7.0
 [0.6.0]: https://github.com/tetherto/qvac-harness/compare/v0.5.0...v0.6.0

package/dist-lib/index.js

@@ -21,7 +21,7 @@ import { buildWavBuffer } from './lib/wav-encode.js';
 import { createVoiceLane, packSamples, speakOnce, speakStreamOnce, transcribeOnce, transcribeStreaming, SUPERTONIC_SAMPLE_RATE } from './lib/capabilities/voice.js';
 import { pushQueue } from './lib/queue.js';
 import { AbortedError, raceAbort } from './lib/helpers/race-abort.js';
-import { audioGenDownloadMetadata, customProjectionMap, defaultProjectionMap, defaultTtsFor, defaultVoiceModel, diffusionDownloadMetadata, curatedModelAsset, infraModelsFor, modelAddonOf, modelNameToCapabilityProfile, musicModelSet, residentSrc, setCustomModels } from './lib/models/model-catalog.js';
+import { audioGenDownloadMetadata, categoryModelConfig, customProjectionMap, defaultProjectionMap, defaultTtsFor, defaultVoiceModel, diffusionDownloadMetadata, curatedModelAsset, infraModelsFor, modelAddonOf, modelNameToCapabilityProfile, musicModelSet, residentSrc, setCustomModels } from './lib/models/model-catalog.js';
 import { createModelStatusCache, customModelRows, invalidateModelStatusCache, mergeCustomRows, listModels } from './lib/models/model-list.js';
 import { deleteModelFiles, summarizeDelete } from './lib/models/model-delete.js';
 import { createEmbedPool, createLoadTracker, createModelPool, createVoicePool, awaitModelCapacity, evictIdle, evictIdleWhere, evictModel, evictModelWithin, isResident, leaseResidentModel, leaseCount, loadedModels, loadingModels, settledModels, modelKey, modelOf, releaseModel, resolveLoadOptions, progressState } from './lib/models/model-pool.js';
@@ -662,6 +662,7 @@ export class Harness extends ReadyResource {
             modelType: 'tts-ggml',
             addon: 'tts',
             modelConfig: TTS_MODEL_CONFIG,
+            modelConfigFor: categoryModelConfig,
             loads: this._loads
         });
     }

package/dist-lib/lib/models/model-catalog.d.ts

@@ -1,6 +1,7 @@
 import type { ModelRegistryEntry, ToolDialect } from '@qvac/sdk';
 import type { ImageModelSet } from '../tools/generate-image.ts';
 import type { MusicModelStages } from '../tools/generate-music.ts';
+import type { ModelSrc } from './context.ts';
 import type { ModelAddon } from './model-source.ts';
 export interface SdkModelAsset extends ModelRegistryEntry {
     readonly src: string;
@@ -22,8 +23,10 @@ export interface CategoryDefault {
     readonly name: string;
     readonly category: ModelCategory;
     readonly asset: SdkModelAsset;
+    readonly modelConfig?: Readonly<Record<string, unknown>>;
 }
 export declare const CATEGORY_DEFAULTS: ReadonlyArray<CategoryDefault>;
+export declare function categoryModelConfig(modelSrc: ModelSrc): Readonly<Record<string, unknown>> | undefined;
 export declare function defaultVoiceModel(category: ModelCategory): string;
 export declare function defaultTtsFor(profile: CapabilityProfile): CategoryDefault;
 export interface CatalogModel {

package/dist-lib/lib/models/model-catalog.js

@@ -1,4 +1,4 @@
-import { AUDIOGEN_ACESTEP_5HZ_LM_0_6B_Q8_0, AUDIOGEN_ACESTEP_V15_SFT_Q8_0, AUDIOGEN_ACESTEP_V15_TURBO_Q4_K_M, AUDIOGEN_ACESTEP_V15_TURBO_Q8_0, AUDIOGEN_QWEN3_EMBEDDING_0_6B_Q8_0, AUDIOGEN_VAE_BF16, EMBEDDINGGEMMA_300M_Q4_0, FLUX_2_KLEIN_4B_Q4_0, FLUX_2_KLEIN_4B_VAE, HEALTHCARE_1_7B_MEDICAL_Q4_K_M, HEALTHCARE_4B_MEDICAL_Q4_K_M, MMPROJ_QWEN3_5_2B_MULTIMODAL_Q8_0, MMPROJ_QWEN3_5_4B_MULTIMODAL_Q8_0, MMPROJ_QWEN3_5_9B_MULTIMODAL_F16, MMPROJ_QWEN3_6_27B_MULTIMODAL_F16, MMPROJ_QWEN3_6_35B_A3B_MULTIMODAL_F16, MMPROJ_QWEN3_8_27B_MULTIMODAL_F16, QWEN3_5_2B_MULTIMODAL_Q4_K_M, QWEN3_5_4B_MULTIMODAL_Q4_K_M, QWEN3_5_9B_MULTIMODAL_Q4_K_M, PARAKEET_TDT_0_6B_V3_Q8_0, QWEN3_4B_Q4_K_M, QWEN3_6_27B_MULTIMODAL_Q4_K_XL, QWEN3_6_35B_A3B_MULTIMODAL_Q4_K_M, QWEN3_8_27B_MULTIMODAL_UD_Q4_K_XL, SD_V2_1_1B_Q8_0, SDXL_BASE_1_0_3B_Q4_0, TTS_MULTILINGUAL_SUPERTONIC3_Q4_0, TTS_MULTILINGUAL_SUPERTONIC3_Q8_0 } from '@qvac/sdk/models';
+import { AUDIOGEN_ACESTEP_5HZ_LM_0_6B_Q8_0, AUDIOGEN_ACESTEP_V15_SFT_Q8_0, AUDIOGEN_ACESTEP_V15_TURBO_Q4_K_M, AUDIOGEN_ACESTEP_V15_TURBO_Q8_0, AUDIOGEN_QWEN3_EMBEDDING_0_6B_Q8_0, AUDIOGEN_VAE_BF16, EMBEDDINGGEMMA_300M_Q4_0, FLUX_2_KLEIN_4B_Q4_0, FLUX_2_KLEIN_4B_VAE, HEALTHCARE_1_7B_MEDICAL_Q4_K_M, HEALTHCARE_4B_MEDICAL_Q4_K_M, MMPROJ_QWEN3_5_2B_MULTIMODAL_Q8_0, MMPROJ_QWEN3_5_4B_MULTIMODAL_Q8_0, MMPROJ_QWEN3_5_9B_MULTIMODAL_F16, MMPROJ_QWEN3_6_27B_MULTIMODAL_F16, MMPROJ_QWEN3_6_35B_A3B_MULTIMODAL_F16, MMPROJ_QWEN3_8_27B_MULTIMODAL_F16, QWEN3_5_2B_MULTIMODAL_Q4_K_M, QWEN3_5_4B_MULTIMODAL_Q4_K_M, QWEN3_5_9B_MULTIMODAL_Q4_K_M, PARAKEET_TDT_0_6B_V3_Q8_0, QWEN3_4B_Q4_K_M, QWEN3_6_27B_MULTIMODAL_Q4_K_XL, QWEN3_6_35B_A3B_MULTIMODAL_Q4_K_M, QWEN3_8_27B_MULTIMODAL_UD_Q4_K_XL, SD_V2_1_1B_Q8_0, SDXL_BASE_1_0_3B_Q4_0, TTS_MULTILINGUAL_SUPERTONIC3_FP16, TTS_MULTILINGUAL_SUPERTONIC3_Q4_0, TTS_MULTILINGUAL_SUPERTONIC3_Q8_0 } from '@qvac/sdk/models';
 import { detectToolDialect } from './tool-dialect.js';
 import { defaultTtsModelName } from './tts-model-default.js';
 import { CAPABILITY_PROFILE_ORDER, QWEN_35_2B, QWEN_35_4B, QWEN_35_9B, QWEN_36_27B, QWEN_36_35B_A3B, QWEN_38_27B, MEDPSY_1_7B, MEDPSY_4B, isCapabilityProfile, meetsMinCapabilityProfile, modelNameToCapabilityProfile, capabilityProfileToModelName, isMultimodalLlm, AGENT_MODEL_NAMES, modelDisplayName, capabilityProfileModelLabel, DEFAULT_IMAGE_MODEL_CHOICE, IMAGE_CHOICE_CATALOG, isMobilePlatform, runnableModelName } from './model-catalog-constants.js';
@@ -62,9 +62,21 @@ export function defaultProjectionMap() {
     }
     return map;
 }
+const TTS_FULL_STEPS = { ttsNumInferenceSteps: 16 };
 export const CATEGORY_DEFAULTS = [
     { name: 'Supertonic3-Q4_0', category: 'tts', asset: TTS_MULTILINGUAL_SUPERTONIC3_Q4_0 },
-    { name: 'Supertonic3-Q8_0', category: 'tts', asset: TTS_MULTILINGUAL_SUPERTONIC3_Q8_0 },
+    {
+        name: 'Supertonic3-Q8_0',
+        category: 'tts',
+        asset: TTS_MULTILINGUAL_SUPERTONIC3_Q8_0,
+        modelConfig: TTS_FULL_STEPS
+    },
+    {
+        name: 'Supertonic3-FP16',
+        category: 'tts',
+        asset: TTS_MULTILINGUAL_SUPERTONIC3_FP16,
+        modelConfig: TTS_FULL_STEPS
+    },
     {
         name: 'Parakeet-TDT-0.6B-v3-Q8_0',
         category: 'transcription',
@@ -72,6 +84,10 @@ export const CATEGORY_DEFAULTS = [
     },
     { name: 'EmbeddingGemma-300M-Q4_0', category: 'embedding', asset: EMBEDDINGGEMMA_300M_Q4_0 }
 ];
+export function categoryModelConfig(modelSrc) {
+    const src = typeof modelSrc === 'string' ? modelSrc : modelSrc.src;
+    return CATEGORY_DEFAULTS.find((entry) => entry.asset.src === src)?.modelConfig;
+}
 export function defaultVoiceModel(category) {
     const entry = CATEGORY_DEFAULTS.find((candidate) => candidate.category === category);
     if (!entry) {

package/dist-lib/lib/models/model-pool.d.ts

@@ -156,8 +156,9 @@ export interface VoicePoolOptions {
     modelType: string;
     addon: ModelAddon;
     modelConfig?: Record<string, unknown> | undefined;
+    modelConfigFor?: (modelSrc: ModelSrc) => Readonly<Record<string, unknown>> | undefined;
     openclCacheDir?: (() => string | undefined) | undefined;
     loads?: LoadTracker;
 }
-export declare function createVoicePool(loadSdk: LoadVoiceSdk, notify: NotifyFn, { modelType, addon, modelConfig, openclCacheDir, loads }: VoicePoolOptions): ModelPuddle;
+export declare function createVoicePool(loadSdk: LoadVoiceSdk, notify: NotifyFn, { modelType, addon, modelConfig, modelConfigFor, openclCacheDir, loads }: VoicePoolOptions): ModelPuddle;
 export {};

package/dist-lib/lib/models/model-pool.js

@@ -392,14 +392,16 @@ export function createEmbedPool(loadSdk, notify, { faults, batchSize = DEFAULT_E
         ...NO_IDLE_GC
     });
 }
-export function createVoicePool(loadSdk, notify, { modelType, addon, modelConfig, openclCacheDir, loads = createLoadTracker() }) {
+export function createVoicePool(loadSdk, notify, { modelType, addon, modelConfig, modelConfigFor, openclCacheDir, loads = createLoadTracker() }) {
     const wantsGpu = modelConfig?.['useGPU'] === true;
     let gpuRefused = false;
     const load = async (model) => {
         const sdk = await loadSdk();
         const modelSrc = await resolveModelSrc(sdk, model, addon);
+        const tuned = modelConfigFor?.(modelSrc);
+        const loadConfig = tuned ? { ...modelConfig, ...tuned } : modelConfig;
         const attempt = () => {
-            const config = gpuRefused ? { ...modelConfig, useGPU: false } : modelConfig;
+            const config = gpuRefused ? { ...loadConfig, useGPU: false } : loadConfig;
             const openclCache = gpuRefused ? undefined : openclCacheDir?.();
             return loads.track(sdk.loadModel({
                 modelSrc,

package/dist-lib/lib/models/tts-model-default.js

@@ -1,9 +1,9 @@
 const TTS_MODEL_BY_PROFILE = {
     tiny: 'Supertonic3-Q4_0',
     low: 'Supertonic3-Q4_0',
-    mid: 'Supertonic3-Q8_0',
-    high: 'Supertonic3-Q8_0',
-    xhigh: 'Supertonic3-Q8_0'
+    mid: 'Supertonic3-FP16',
+    high: 'Supertonic3-FP16',
+    xhigh: 'Supertonic3-FP16'
 };
 export const TTS_MODEL_NAMES = [
     ...new Set(Object.values(TTS_MODEL_BY_PROFILE))

package/dist-lib/lib/tools/web-fetch.js

@@ -32,7 +32,8 @@ export function createWebFetchTool({ fetch: fetchImpl, resolveHost = lookupHost
             'shares, when a search snippet is not enough to answer. ' +
             'When the URL serves a file rather than a page — PDF, .docx, .xlsx, .pptx, an image, any ' +
             'binary — this is also the tool that downloads it: the bytes become a chat attachment and ' +
-            'the result carries its attachmentId. It is the only way to get a remote file into the ' +
+            'the result carries its attachmentId. A plain-text file (.txt, .csv, .json…) is attached ' +
+            'too, and its text is also returned. It is the only way to get a remote file into the ' +
             'chat; exec has no network and http_request truncates binary bodies.',
         parameters: {
             type: 'object',
@@ -133,13 +134,22 @@ export function createWebFetchTool({ fetch: fetchImpl, resolveHost = lookupHost
                 const imagesLine = attached.length
                     ? `\n\n[Attached ${attached.length === 1 ? 'image' : 'images'}: ${attached.join(', ')} — shown to the user automatically]`
                     : '';
+                const file = isHtml
+                    ? null
+                    : await attachTextFile(response, { url: currentUrl, mimeType, attach: ctx?.attach });
+                const fileLine = file
+                    ? `\n\n[Attached the file "${file.fileName}" (${Math.round(file.byteLength / 1024)}KB) — the user can open or save it; stage it with exec as inputs [{ "attachmentId": "${file.id}", "path": "${file.fileName}" }] to read past the text above]`
+                    : '';
                 const video = isHtml ? pageVideoRef(body, currentUrl) : null;
                 const videoLine = video
                     ? `\n\n[This page's video: ${video.url} — share this link when the user wants to watch it]`
                     : '';
                 const header = page.title ? `${page.title}\nSource: ${url}` : `Source: ${url}`;
                 return {
-                    content: `${header}\n\n${text}${imagesLine}${videoLine}`,
+                    attachmentId: file?.id,
+                    fileName: file?.fileName,
+                    byteLength: file?.byteLength,
+                    content: `${header}\n\n${text}${fileLine}${imagesLine}${videoLine}`,
                     data: {
                         results: [
                             {
@@ -201,6 +211,30 @@ async function attachDocument(response, { url, mimeType, attach }) {
             'user to attach a file this call already attached.'
     };
 }
+async function attachTextFile(response, { url, mimeType, attach }) {
+    if (!attach)
+        return null;
+    const bytes = await response.bytes?.();
+    if (!bytes || bytes.byteLength === 0 || bytes.byteLength >= MAX_PAGE_BYTES) {
+        return null;
+    }
+    const mime = mimeType || 'text/plain';
+    try {
+        return await attach({
+            fileName: attachmentFileName(url, mime),
+            mimeType: mime,
+            bytes,
+            raw: true
+        });
+    }
+    catch {
+        return null;
+    }
+}
+const TEXT_EXTENSIONS = {
+    'text/plain': 'txt',
+    'text/markdown': 'md'
+};
 const MIME_EXTENSIONS = {
     'application/pdf': 'pdf',
     'application/msword': 'doc',
@@ -234,7 +268,8 @@ function attachmentFileName(url, mimeType) {
     if (extension && acceptableExtensions(mimeType).has(extension)) {
         return base;
     }
-    if (mimeType === 'application/octet-stream' && /^[a-z][\w-]*$/.test(extension)) {
+    const trustsUrl = mimeType === 'application/octet-stream' || !isBinaryMime(mimeType);
+    if (trustsUrl && /^[a-z][\w-]*$/.test(extension)) {
         return base;
     }
     return `${base}.${extensionFor(mimeType)}`;
@@ -248,7 +283,7 @@ function acceptableExtensions(mimeType) {
     return accepted;
 }
 function extensionFor(mimeType) {
-    const known = MIME_EXTENSIONS[mimeType];
+    const known = MIME_EXTENSIONS[mimeType] ?? TEXT_EXTENSIONS[mimeType];
     if (known) {
         return known;
     }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@qvac/harness",
-  "version": "0.7.1",
+  "version": "0.7.2",
   "type": "module",
   "engines": {
     "bare": ">=1.34.0"