@qvac/harness 0.7.0 → 0.7.2

package/CHANGELOG.md

@@ -8,6 +8,68 @@ 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
+
+- `list()` rows carry `isLoading`: true from the moment a load of that model
+  starts, whoever asked for it, until it is ready or the load fails. It is read
+  off the pool, so a host shows a model as loading without counting loads
+  itself. A model loading again under other settings while a copy of it is
+  resident reads `isLoaded` and `isLoading` both true. Chat models read it
+  today: image, music and translation models load for each generation, and
+  speech, transcription and embedding pools key by source, so those rows read
+  `false` for now. An optional field on the wire; a row from an older harness
+  reads `false`. (#143)
+
+  ```js
+  const loading = (await harness.list({ cachedOnly: true })).filter((row) => row.isLoading)
+  ```
+
+### Changed
+
+- The `model` event carries `catalogChanged` at every edge a `list()` re-read
+  would show: the `loading` that starts a load and the `ready` or `error` that
+  ends it, beside `unloaded`. Progress ticks still do not. A host that re-reads
+  the list on `catalogChanged` sees a model turn loading, then loaded, with no
+  bookkeeping of its own. The served event stream never drops a frame that
+  carries it, so two `ready` frames of one model under two keys both cross. (#143)
+- `ModelInfo` requires `isLoading`, and `listModels`, `customModelRows` and
+  `mergeCustomRows` return `CatalogRow`, the row before the pools are asked. A
+  typed fake `list()` adds the field. (#143)
+- A custom model's row reads `isLoaded` only once its load has landed; it used
+  to read `true` from the first attempt. (#143)
+
+### Fixed
+
+- A load that ends without the model in memory emits `model` with `error` and
+  the reason: it failed, or a suspend or close cancelled it. It used to end in
+  silence, so a host could not tell a failed load from one still running. The
+  `error` also refreshes the cached `isCached`, since a load may have downloaded
+  the weights before it failed. (#143)
+
 ## [0.7.0] - 2026-09-29
 
 ### Added
@@ -979,7 +1041,9 @@ 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/v0.7.0...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
 [0.5.0]: https://github.com/tetherto/qvac-harness/compare/v0.4.0...v0.5.0

package/README.md

@@ -165,6 +165,7 @@ const custom = models.filter((model) => model.registrySource === 'custom')
 //   registrySource: 'custom',
 //   isCached: false,
 //   isLoaded: false,
+//   isLoading: false,
 //   supportsImages: false,
 //   expectedSize: 1868506272,
 //   label: 'Nanbeige 4.2 3B',
@@ -234,8 +235,20 @@ if (big.reason === 'memory') showNotice('Unload a model first')
 `unloadModel` waits for any request holding the model to finish, then unloads
 every copy of it. `unloaded` is false when a request still held it after the
 wait, so a caller that wants the model gone at once stops those requests first.
-Both change `isLoaded` in the next `list()`, and emit a `model` event with
-`catalogChanged` set. A load or an unload counts as a call in flight for
+Both change `isLoaded` in the next `list()`. `isLoading` is true on a row from
+the moment a load of its model starts, whoever asked for it (a request,
+`warmModel` or `loadModel`), until the model is ready or the load fails; it is
+read off the pool, so a host keeps no count of its own. A model loading again
+under other settings while a copy of it is resident reads `isLoaded: true` and
+`isLoading: true` at once. The `model` event carries `catalogChanged` at each
+edge a `list()` re-read would show: the `loading` that starts a load, the
+`ready` or `error` that ends it, and `unloaded`. Progress ticks between them do
+not. A load that ends without the model in memory reports
+`error` with the reason in `message`: it failed, or a suspend or close cancelled it.
+An image, music or translation model loads for each generation and never reads
+`isLoading: true`; a speech, transcription or embedding model loads under its
+source rather than its row name, so its row reads `false` for now. A load or an
+unload counts as a call in flight for
 [`setModelIdleTimeout`](#freeing-memory-when-idle), and one that lands during an
 idle unload waits for it, so a load never loads a second copy and an unload
 answers once the memory is free.

package/dist-lib/index.d.ts

@@ -218,6 +218,9 @@ export declare class Harness extends ReadyResource {
     describe(): Promise<HarnessInfo>;
     diagnostics(): HarnessDiagnostics;
     private _loadedModels;
+    private _loadingModels;
+    private _settledModels;
+    private _pools;
     private _loadPluginSdk;
     private _loadDiffusionSdk;
     private _slmTranslate;

package/dist-lib/index.js

@@ -21,10 +21,10 @@ 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, modelKey, modelOf, releaseModel, resolveLoadOptions, progressState } from './lib/models/model-pool.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';
 import { defaultContextSize } from './lib/models/context.js';
 import { resolveMainGpu } from './lib/models/gpu-selection.js';
 import { createIdleReaper } from './lib/models/idle-unload.js';
@@ -558,13 +558,22 @@ export class Harness extends ReadyResource {
         };
     }
     _loadedModels() {
-        const pools = [
+        return this._pools().flatMap((pool) => loadedModels(pool));
+    }
+    _loadingModels() {
+        return new Set(this._pools().flatMap((pool) => loadingModels(pool)));
+    }
+    _settledModels() {
+        return new Set(this._pools().flatMap((pool) => settledModels(pool)));
+    }
+    _pools() {
+        return [
+            this._models,
             this._embedModels,
             ...this._retiredEmbedModels,
             this._ttsModels,
             this._transcriptionModels
         ];
-        return [...loadedModels(this._models), ...pools.flatMap((pool) => loadedModels(pool))];
     }
     _loadPluginSdk(current, store, importPlugin) {
         let pending = current();
@@ -653,6 +662,7 @@ export class Harness extends ReadyResource {
             modelType: 'tts-ggml',
             addon: 'tts',
             modelConfig: TTS_MODEL_CONFIG,
+            modelConfigFor: categoryModelConfig,
             loads: this._loads
         });
     }
@@ -667,8 +677,7 @@ export class Harness extends ReadyResource {
         });
     }
     _notifyModel(state) {
-        const catalogChanged = state.status === 'ready' || state.status === 'unloaded';
-        if (catalogChanged) {
+        if (state.status === 'ready' || state.status === 'unloaded' || state.status === 'error') {
             invalidateModelStatusCache(this._modelStatus);
         }
         if (state.status === 'unloaded') {
@@ -678,7 +687,11 @@ export class Harness extends ReadyResource {
                 }
             }
         }
-        this.emit('model', { ...state, catalogChanged, loaded: this._loadedModels() });
+        this.emit('model', {
+            ...state,
+            catalogChanged: state.catalogChanged === true,
+            loaded: this._loadedModels()
+        });
     }
     async _collectRetiredEmbedPool(pool) {
         if (!this._retiredEmbedModels.has(pool) || pool.idle !== pool.size)
@@ -1055,17 +1068,17 @@ export class Harness extends ReadyResource {
         if (!this.opened)
             await this.ready();
         const sdk = await this._sdk.load();
-        const [rows, custom] = await Promise.all([
-            listModels(sdk, options, {
-                cache: this._modelStatus,
-                projectionFor: (name) => this._projectionFor(name)
-            }),
-            customModelRows(options, {
-                modelsDir: this._modelsDir,
-                loaded: new Set(this._loadedModels().map((entry) => entry.model))
-            })
-        ]);
-        return mergeCustomRows(rows, custom);
+        const rows = await listModels(sdk, options, {
+            cache: this._modelStatus,
+            projectionFor: (name) => this._projectionFor(name)
+        });
+        const loaded = this._settledModels();
+        const loading = this._loadingModels();
+        const custom = await customModelRows(options, { modelsDir: this._modelsDir, loaded });
+        return mergeCustomRows(rows, custom).map((row) => ({
+            ...row,
+            isLoading: loading.has(row.name)
+        }));
     }
     async deleteModel(name) {
         if (!this.opened)

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-handle.d.ts

@@ -5,13 +5,17 @@ export interface ModelResource {
 }
 export interface ModelHandleOptions {
     load(): Promise<ModelResource>;
+    onLoaded?(): void;
+    onFailed?(err: unknown): void;
 }
 export declare class ModelHandle {
     private readonly _load;
     private readonly _peek;
+    private readonly _onLoaded;
+    private readonly _onFailed;
     private _tracked;
     private _loading;
-    constructor({ load }: ModelHandleOptions);
+    constructor({ load, onLoaded, onFailed }: ModelHandleOptions);
     load(): Promise<ModelResource>;
     get loading(): boolean;
     peek(): Promise<ModelResource> | null;

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

@@ -2,12 +2,16 @@ import { retryMemo } from '../helpers/retry-memo.js';
 export class ModelHandle {
     _load;
     _peek;
+    _onLoaded;
+    _onFailed;
     _tracked = null;
     _loading = false;
-    constructor({ load }) {
+    constructor({ load, onLoaded, onFailed }) {
         const memo = retryMemo(load);
         this._load = memo;
         this._peek = memo.peek;
+        this._onLoaded = onLoaded;
+        this._onFailed = onFailed;
         this.load().catch(noop);
     }
     load() {
@@ -15,12 +19,13 @@ export class ModelHandle {
         if (attempt !== this._tracked) {
             this._tracked = attempt;
             this._loading = true;
-            const settled = () => {
+            const settle = (report) => {
                 if (this._tracked === attempt) {
                     this._loading = false;
+                    report();
                 }
             };
-            attempt.then(settled, settled);
+            attempt.then(() => settle(() => this._onLoaded?.()), (err) => settle(() => this._onFailed?.(err)));
         }
         return attempt;
     }

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

@@ -16,12 +16,14 @@ export interface ModelInfo {
     registrySource: string;
     isCached: boolean;
     isLoaded: boolean;
+    isLoading: boolean;
     supportsImages: boolean;
     expectedSize: number;
     actualSize?: number;
     label?: string;
     taskType: ModelTaskType;
 }
+export type CatalogRow = Omit<ModelInfo, 'isLoading'>;
 export interface ListModelsOptions {
     type?: ModelType;
     cachedOnly?: boolean;
@@ -45,11 +47,11 @@ export interface ListModelsRuntime {
     cache?: ModelStatusCache | undefined;
     projectionFor?: ((name: string) => string | undefined) | undefined;
 }
-export declare function listModels(sdk: ModelListSdk, { type, cachedOnly, fullRegistry }?: ListModelsOptions, { cache, projectionFor }?: ListModelsRuntime): Promise<ModelInfo[]>;
+export declare function listModels(sdk: ModelListSdk, { type, cachedOnly, fullRegistry }?: ListModelsOptions, { cache, projectionFor }?: ListModelsRuntime): Promise<CatalogRow[]>;
 export interface CustomModelRowsContext {
     readonly modelsDir?: string;
     readonly loaded: ReadonlySet<string>;
 }
-export declare function customModelRows({ type, cachedOnly }: ListModelsOptions, { modelsDir, loaded }: CustomModelRowsContext): Promise<ModelInfo[]>;
-export declare function mergeCustomRows(catalog: ReadonlyArray<ModelInfo>, custom: ReadonlyArray<ModelInfo>): ModelInfo[];
+export declare function customModelRows({ type, cachedOnly }: ListModelsOptions, { modelsDir, loaded }: CustomModelRowsContext): Promise<CatalogRow[]>;
+export declare function mergeCustomRows(catalog: ReadonlyArray<CatalogRow>, custom: ReadonlyArray<CatalogRow>): CatalogRow[];
 export {};

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

@@ -5,8 +5,9 @@ import { type ModelAddon } from './model-source.ts';
 import type { GpuFaultState } from './sdk-faults.ts';
 type SdkModule = typeof import('@qvac/sdk');
 export interface ModelState {
-    status: 'loading' | 'downloading' | 'downloaded' | 'ready' | 'unloading' | 'unloaded';
+    status: 'loading' | 'downloading' | 'downloaded' | 'ready' | 'error' | 'unloading' | 'unloaded';
     model: string;
+    catalogChanged?: boolean;
     [key: string]: unknown;
 }
 export type NotifyFn = (state: ModelState) => void;
@@ -78,6 +79,8 @@ type PoolLike = ModelPuddle;
 export declare function modelOf(key: string): string;
 export declare function releaseModel(pool: PoolLike, key: string): void;
 export declare function loadedModels(pool: PoolLike): LoadedModel[];
+export declare function settledModels(pool: PoolLike): string[];
+export declare function loadingModels(pool: PoolLike): string[];
 export interface ResidentLease {
     model: string;
     modelId: string;
@@ -153,8 +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

@@ -10,6 +10,7 @@ import { hlog } from '../log.js';
 import { RUNAWAY_STOP_SEQUENCES } from '../helpers/control-tokens.js';
 import { raceTimeout } from '../helpers/race-timeout.js';
 import { resolveModelSrc } from './model-source.js';
+import { errorMessage } from '../guards.js';
 export function createLoadTracker() {
     const inflight = new Set();
     let draining = null;
@@ -101,6 +102,34 @@ export function loadedModels(pool) {
     }
     return out;
 }
+export function settledModels(pool) {
+    const out = [];
+    for (const entry of pool.resources.values()) {
+        if (entry.value.peek() !== null && !entry.value.loading) {
+            out.push(modelOf(entry.id));
+        }
+    }
+    return out;
+}
+export function loadingModels(pool) {
+    const out = [];
+    for (const entry of pool.resources.values()) {
+        if (entry.value.loading) {
+            out.push(modelOf(entry.id));
+        }
+    }
+    return out;
+}
+function reportedHandle(notify, model, load) {
+    return new ModelHandle({
+        load: () => {
+            notify({ status: 'loading', model, catalogChanged: true });
+            return load();
+        },
+        onLoaded: () => notify({ status: 'ready', model, catalogChanged: true }),
+        onFailed: (err) => notify({ status: 'error', model, message: errorMessage(err), catalogChanged: true })
+    });
+}
 export async function leaseResidentModel(pool, wanted = () => true) {
     for (const entry of pool.resources.values()) {
         const key = entry.id;
@@ -292,7 +321,6 @@ export function loadClockGuard(model, sdk, { now = () => Date.now(), monoMs = mo
 export function createModelPool(loadSdk, notify, { deviceConfig = {}, loads = createLoadTracker() } = {}) {
     const load = async (model, opts) => {
         const sdk = await loadSdk();
-        notify({ status: 'loading', model });
         const modelSrc = await resolveModelSrc(sdk, model);
         const resolved = resolveLoadOptions(opts, deviceConfig);
         const ctxSize = resolved.ctxSize || defaultContextSize(modelSrc, customContextLength(model));
@@ -332,11 +360,10 @@ export function createModelPool(loadSdk, notify, { deviceConfig = {}, loads = cr
             onProgress: (progress) => notify(progressState(model, progress))
         })));
         hlog('llm loaded', { model, loadMs: Date.now() - loadStarted });
-        notify({ status: 'ready', model });
         return { model, modelId, ctxSize };
     };
     return new ModelPuddle({
-        open: (key, model = key, opts = {}) => new ModelHandle({ load: () => load(model, opts) }),
+        open: (key, model = key, opts = {}) => reportedHandle(notify, model, () => load(model, opts)),
         gc: unloadOnGc(loadSdk, notify),
         ...NO_IDLE_GC
     });
@@ -344,7 +371,6 @@ export function createModelPool(loadSdk, notify, { deviceConfig = {}, loads = cr
 export function createEmbedPool(loadSdk, notify, { faults, batchSize = DEFAULT_EMBED_BATCH_SIZE, openclCacheDir, loads = createLoadTracker() }) {
     const load = async (model) => {
         const sdk = await loadSdk();
-        notify({ status: 'loading', model });
         const modelSrc = await resolveModelSrc(sdk, model, 'embeddings');
         const openclCache = faults.device !== 'cpu' && openclCacheDir ? openclCacheDir() : undefined;
         const modelId = await loadClockGuard(model, sdk).race(loads.track(sdk.loadModel({
@@ -358,24 +384,24 @@ export function createEmbedPool(loadSdk, notify, { faults, batchSize = DEFAULT_E
             },
             onProgress: (progress) => notify(progressState(model, progress))
         })));
-        notify({ status: 'ready', model });
         return { model, modelId };
     };
     return new ModelPuddle({
-        open: (key, model = key) => new ModelHandle({ load: () => load(model) }),
+        open: (key, model = key) => reportedHandle(notify, model, () => load(model)),
         gc: unloadOnGc(loadSdk, notify),
         ...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();
-        notify({ status: 'loading', model });
         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,
@@ -391,11 +417,10 @@ export function createVoicePool(loadSdk, notify, { modelType, addon, modelConfig
             gpuRefused = true;
             return attempt();
         }));
-        notify({ status: 'ready', model });
         return { model, modelId };
     };
     return new ModelPuddle({
-        open: (key, model = key) => new ModelHandle({ load: () => load(model) }),
+        open: (key, model = key) => reportedHandle(notify, model, () => load(model)),
         gc: unloadOnGc(loadSdk, notify),
         ...NO_IDLE_GC
     });
@@ -423,7 +448,7 @@ function unloadOnGc(loadSdk, notify) {
             catch {
                 continue;
             }
-            notify({ status: 'unloaded', model });
+            notify({ status: 'unloaded', model, catalogChanged: true });
         }
     };
 }

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/serve.js

@@ -270,7 +270,7 @@ function modelEvents(harness) {
             catalogChanged,
             ...loaded.map((entry) => `${entry.model}:${entry.refs}`)
         ].join('\0');
-        if (key === last) {
+        if (key === last && !catalogChanged) {
             return;
         }
         last = key;

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/dist-lib/schema/harness.js

@@ -339,7 +339,8 @@ export function registerHarnessTypes(ns) {
             { name: 'supportsImages', type: 'bool', required: true },
             { name: 'expectedSize', type: 'uint', required: true },
             { name: 'actualSize', type: 'uint' },
-            { name: 'label', type: 'string' }
+            { name: 'label', type: 'string' },
+            { name: 'isLoading', type: 'bool' }
         ]
     });
     ns.register({

package/dist-lib/spec/hrpc/messages.js

@@ -1,12 +1,12 @@
 // This file is autogenerated by the hyperschema compiler
-// Schema Version: 17
+// Schema Version: 18
 /* eslint-disable camelcase */
 /* eslint-disable quotes */
 /* eslint-disable space-before-function-paren */
 
 import { c } from 'hyperschema/runtime'
 
-const VERSION = 17
+const VERSION = 18
 
 // eslint-disable-next-line no-unused-vars
 let version = VERSION
@@ -1610,7 +1610,7 @@ const encoding31 = {
     c.string.preencode(state, m.type)
     c.string.preencode(state, m.registryPath)
     c.string.preencode(state, m.registrySource)
-    state.end++ // max flag is 16 so always one byte
+    state.end++ // max flag is 32 so always one byte
     c.uint.preencode(state, m.expectedSize)
 
     if (m.actualSize) c.uint.preencode(state, m.actualSize)
@@ -1622,7 +1622,8 @@ const encoding31 = {
       (m.isLoaded ? 2 : 0) |
       (m.supportsImages ? 4 : 0) |
       (m.actualSize ? 8 : 0) |
-      ((version >= 15 && m.label) ? 16 : 0)
+      ((version >= 15 && m.label) ? 16 : 0) |
+      ((version >= 18 && m.isLoading) ? 32 : 0)
 
     c.string.encode(state, m.name)
     c.string.encode(state, m.type)
@@ -1651,7 +1652,8 @@ const encoding31 = {
       supportsImages: (flags & 4) !== 0,
       expectedSize: c.uint.decode(state),
       actualSize: (flags & 8) !== 0 ? c.uint.decode(state) : 0,
-      label: (version >= 15 && (flags & 16) !== 0) ? c.string.decode(state) : null
+      label: (version >= 15 && (flags & 16) !== 0) ? c.string.decode(state) : null,
+      isLoading: (version >= 18 && (flags & 32) !== 0)
     }
   }
 }

package/dist-lib/spec/hyperschema/index.js

@@ -1,12 +1,12 @@
 // This file is autogenerated by the hyperschema compiler
-// Schema Version: 17
+// Schema Version: 18
 /* eslint-disable camelcase */
 /* eslint-disable quotes */
 /* eslint-disable space-before-function-paren */
 
 import { c } from 'hyperschema/runtime'
 
-const VERSION = 17
+const VERSION = 18
 
 // eslint-disable-next-line no-unused-vars
 let version = VERSION
@@ -1610,7 +1610,7 @@ const encoding31 = {
     c.string.preencode(state, m.type)
     c.string.preencode(state, m.registryPath)
     c.string.preencode(state, m.registrySource)
-    state.end++ // max flag is 16 so always one byte
+    state.end++ // max flag is 32 so always one byte
     c.uint.preencode(state, m.expectedSize)
 
     if (m.actualSize) c.uint.preencode(state, m.actualSize)
@@ -1622,7 +1622,8 @@ const encoding31 = {
       (m.isLoaded ? 2 : 0) |
       (m.supportsImages ? 4 : 0) |
       (m.actualSize ? 8 : 0) |
-      ((version >= 15 && m.label) ? 16 : 0)
+      ((version >= 15 && m.label) ? 16 : 0) |
+      ((version >= 18 && m.isLoading) ? 32 : 0)
 
     c.string.encode(state, m.name)
     c.string.encode(state, m.type)
@@ -1651,7 +1652,8 @@ const encoding31 = {
       supportsImages: (flags & 4) !== 0,
       expectedSize: c.uint.decode(state),
       actualSize: (flags & 8) !== 0 ? c.uint.decode(state) : 0,
-      label: (version >= 15 && (flags & 16) !== 0) ? c.string.decode(state) : null
+      label: (version >= 15 && (flags & 16) !== 0) ? c.string.decode(state) : null,
+      isLoading: (version >= 18 && (flags & 32) !== 0)
     }
   }
 }

package/dist-lib/spec/hyperschema/schema.json

@@ -1,5 +1,5 @@
 {
-  "version": 17,
+  "version": 18,
   "schema": [
     {
       "name": "empty",
@@ -1341,6 +1341,11 @@
           "name": "label",
           "type": "string",
           "version": 15
+        },
+        {
+          "name": "isLoading",
+          "type": "bool",
+          "version": 18
         }
       ]
     },

package/dist-lib/spec/hyperschema/types.d.ts

@@ -1,5 +1,5 @@
 // This file is autogenerated by hyperschema-ts
-// Schema version: 17
+// Schema version: 18
 // Do not edit manually.
 
 export interface HarnessEmpty {
@@ -257,6 +257,7 @@ export interface HarnessModelInfo {
   expectedSize: number
   actualSize?: number
   label?: string | null
+  isLoading?: boolean
 }
 
 export interface HarnessListModelsRequest {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@qvac/harness",
-  "version": "0.7.0",
+  "version": "0.7.2",
   "type": "module",
   "engines": {
     "bare": ">=1.34.0"