Release 0.1.74 [auto-version-bump]

.gitea/workflows/publish.yaml

@@ -29,22 +29,14 @@ jobs:
           python3 scripts/bump-version.py --check
           scripts/check-frontend.sh
 
-      - name: Build Piper WebAssembly assets
+      - name: Bundle Piper worker client code
         shell: bash
         run: |
           set -euo pipefail
           npm install --no-audit --no-fund
           npm run build:piper
-          test -s public/tts/piper-wasm-v1/piper/piper_phonemize.wasm
-          test -s public/tts/piper-wasm-v1/piper/piper_phonemize.data
           test -s public/tts/piper-wasm-v1/piper-worker.js
           test -s public/piper-runtime.js
-          test -s public/tts/piper-wasm-v1/voices/sv/sv_SE/alma/medium/sv_SE-alma-medium.onnx
-          test -s public/tts/piper-wasm-v1/voices/sv/sv_SE/alma/medium/sv_SE-alma-medium.onnx.json
-          test -s public/tts/piper-wasm-v1/voices/sv/sv_SE/lisa/medium/sv_SE-lisa-medium.onnx
-          test -s public/tts/piper-wasm-v1/voices/sv/sv_SE/lisa/medium/sv_SE-lisa-medium.onnx.json
-          test -s public/tts/piper-wasm-v1/voices/sv/sv_SE/nst/medium/sv_SE-nst-medium.onnx
-          test -s public/tts/piper-wasm-v1/voices/sv/sv_SE/nst/medium/sv_SE-nst-medium.onnx.json
 
       - name: Ensure Helm is available
         shell: bash

AGENTS.md

@@ -3,9 +3,12 @@
 - This repository owns only the static, accessible web client and its container/chart packaging.
 - Rust remains the sole authority for game state. The frontend calls the backend through same-origin `/api/...` and `/version` paths.
 - Keep frontend and backend behind `ai-spel.k3s.k0.nu` so session cookies remain same-origin.
+- The public footer displays frontend, static-service, and API versions; load static-service version from uncached `/static/version.json`.
 - `VERSION` is the only release-version source in this repository. Use `scripts/bump-version.py`; do not manually edit chart versions.
 - Changes are written locally and built, checked, published, and exercised by Gitea Actions on the existing `large` runner and Kubernetes. Never run apps, tests, containers, images, or binaries locally. Do not create or register runners.
 - Never commit secrets, generated credentials, kubeconfigs, or environment files.
 - Helm charts are published to `harbor.k3s.k0.nu/ai-spel`; Argo CD uses the existing `ai-spel` project and namespace.
-- Player-facing speech uses the local Piper WebAssembly adapter when the player has activated audio. All supported Swedish Piper voice models are packaged into the frontend image and served from this origin; model downloads are redirected to same-origin paths and cached by the browser when supported. Game text is synthesized on-device and must never be sent to a TTS provider.
+- Player-facing speech uses the local Piper WebAssembly adapter when the player has activated audio. The separate `ai-spel-static` Nginx service serves pinned Swedish Piper models and ONNX/Piper runtime files from same-origin `/static/piper-wasm-v1/` URLs with immutable caching. Game text is synthesized on-device and must never be sent to a TTS provider.
-- Build Piper/ONNX WebAssembly files and fetch/checksum pinned Swedish model files in the `large` runner; serve them from the frontend image under `/tts/piper-wasm-v1/`. Never build or run these assets locally.
+- This pipeline bundles only the Piper worker/client glue. Model/runtime downloads, image builds, and deployment belong to `ai-spel/ai-spel-static` and the existing `large` runner. Never build or run these assets locally.
+- Browser local AI is paused by default behind `FEATURE_FLAGS.localAI` in `index.html`. When disabled, do not load its adapter or any Wllama/WebAssembly/Gemma assets. If it is deliberately enabled in a later change, it must load the adapter, runtime, model, manifest, YAML prompt, and license resources only from the same-origin `/static/wllama-gemma3-v2/` path owned by `ai-spel/ai-spel-static`. Do not add browser CDN, npm, or model-hub fallbacks. The player must be shown and accept applicable model terms before the model downloads or runs. Acceptance, model failure, or local-AI opt-out must never block play. Incompatible asset changes require a new immutable static path and a coordinated frontend release.
+- Local model output is advisory only. Preserve the original player input, send local hints only as untrusted context to Rust, and label offline replies as unsaved. Local diagnostic logging is opt-in, browser-local, exportable/deletable, and expires after 30 days.

PIPER_VOICE_NOTICES.txt

@@ -1,25 +0,0 @@
-Swedish Piper voice models
-==========================
-
-This frontend distributes the Swedish medium Piper models Alma, Lisa, and NST
-from the rhasspy/piper-voices repository, pinned to revision
-c10ece1aade47bb51c153c893d14e5bf8e5b7117. Model weights are delivered by this
-site from /tts/piper-wasm-v1/voices/ after integrity verification in CI.
-
-Alma (sv_SE-alma-medium)
-Source: https://huggingface.co/rhasspy/piper-voices/tree/c10ece1aade47bb51c153c893d14e5bf8e5b7117/sv/sv_SE/alma/medium
-The model card states CC BY 4.0 and identifies the upstream voice project at
-https://huggingface.co/yeagersthlm/piper-voice-sv-alma.
-
-Lisa (sv_SE-lisa-medium)
-Source: https://huggingface.co/rhasspy/piper-voices/tree/c10ece1aade47bb51c153c893d14e5bf8e5b7117/sv/sv_SE/lisa/medium
-The upstream model card describes this as fine-tuned from the Norwegian
-talesyntese medium voice.
-
-NST (sv_SE-nst-medium)
-Source: https://huggingface.co/rhasspy/piper-voices/tree/c10ece1aade47bb51c153c893d14e5bf8e5b7117/sv/sv_SE/nst/medium
-The upstream model card identifies the NST Swedish TTS dataset as CC0 and
-credits KBLab at the National Library of Sweden for training the voice.
-
-The upstream model files and their model cards remain the source of truth for
-license terms. The complete source revision is pinned in scripts/build-piper.mjs.

README.md

@@ -6,15 +6,17 @@ After login, character creation and play use one dialogue history and one text i
 
 During play the UI has no persistent location, task, inventory, economy, status, or purchase panels. The player asks for information or writes an action in the same input. Rust validates state changes; the AI can only propose a typed intent. A direct request to travel may establish a new place and an encounter when Rust commits that movement.
 
+The dialogue also offers **Gör om senaste** and **Torrkör senaste**. Both resend the latest player action while leaving the previous dialogue turn out of the AI conversation context. “Gör om” processes a fresh, normal turn. “Torrkör” calls the narrator, generates and validates a temporary Luau rule when needed, and evaluates either that temporary rule or an approved stored rule. It rolls back the complete game transaction and does not save the player turn, AI trace, generated rule, event, ownership, or world progress. The reply identifies which rule was evaluated and marks a temporary rule as such.
+
 If the player asks to find or approach someone and no NPC is present, the direct dialogue scene may propose one person and an opening line. Rust creates that NPC only for the explicit request and commits the first exchange idempotently; a later turn continues the saved conversation. Clear requests about the current location are answered from its saved facts without asking the player to restate what they want to inspect.
 
-The separate `/admin` route provides an authenticated operations view for player search/profile/trace, YAML prompt assets, generated Luau rules with validation results, runtime debug logging, and the append-only audit log. All privileged data comes from the Rust `/api/admin/...` API; the browser never connects directly to PostgreSQL. Backend endpoints require `accounts.is_admin=true`. The Driftloggar view lets an administrator switch `debug=true` on or off without restarting pods; the setting is stored in PostgreSQL and propagated to running replicas. Verbose pod logs include request, AI-call, validation, and commit metadata but never raw player text, prompts, model responses, sessions, or secrets. Full protected prompt/provider exchanges remain in the audited player trace.
+The separate `/admin` route provides an authenticated operations view for player search/profile/trace, NPC selection and editing, YAML prompt assets, generated Luau rules with validation results, runtime debug logging, and the append-only audit log. NPC editing covers identity and voice, typed profile fields plus future JSON properties, existing memory reconstructions, and owned items. The UI displays current position and commitments for context. Every NPC read and write goes through Rust, checks `accounts.is_admin=true`, validates domain limits, and writes an audit record in the same transaction as each mutation; the browser never connects directly to PostgreSQL. Profile and memory writes use expected-current values to detect stale edits. The Driftloggar view lets an administrator switch `debug=true` on or off without restarting pods; the setting is stored in PostgreSQL and propagated to running replicas. Verbose pod logs include request, AI-call, validation, and commit metadata but never raw player text, prompts, model responses, sessions, or secrets. Full protected prompt/provider exchanges remain in the audited player trace.
 
-The game is text- and dialogue-based and uses no images. Place descriptions include sound, smell, air/temperature, touch/surface, and spatial-orientation cues. Visible details may also be described in words, but are never required to understand the scene or play. The transcript is screen-reader announced, and Enter submits the input while Shift+Enter inserts a newline; no separate send button is required.
+The game is text- and dialogue-based and uses no images. Place descriptions include sound, smell, air/temperature, touch/surface, and spatial-orientation cues. Visible details may also be described in words, but are never required to understand the scene or play. The transcript is screen-reader announced. A clearly labeled **Skicka** button beside the microphone submits typed text; Enter also submits while Shift+Enter inserts a newline. Typed text, speech transcriptions, and the send button all use the same submission flow.
 
 Player-facing speech uses Piper TTS in WebAssembly in the browser after audio is activated by a user gesture. It reads game dialogue, onboarding, account responses, character and world creation, gameplay replies, in-game errors, and microphone disclosures; the `/admin` route is excluded. The audio is decoded and played through the Web Audio API, independently of `speechSynthesis`. The first unauthenticated visit keeps the introduction visible and it can be started with `Läs introduktionen`. New game replies are read automatically and the transcript remains available to screen readers and as text. Returning players hear the latest saved game reply when they reopen the game. The player can change speaking rate; narrator pitch is neutral and character pitch is varied consistently by voice identity. These preferences are stored in browser `localStorage`. If WebAssembly, model loading, or audio playback fails, the same content remains available as text and the dialogue offers a retry.
 
-The Piper engine, WebAssembly runtime, and all three Swedish medium voice models (Alma, Lisa, and NST) are served from this frontend's same-origin `/tts/piper-wasm-v1/` path. The Gitea pipeline downloads the pinned model files, verifies their sizes and cryptographic hashes, and packages them into the versioned frontend image. The browser's Piper adapter redirects the library's built-in model URLs to this same-origin server; it does not request voice files from Hugging Face. A voice model is fetched on first use and cached in browser storage when supported. The game currently uses the NST model. Dialogue text is synthesized on the device and is not sent to a TTS provider. Voice model attribution and license references are available in the voice settings and at `/tts/piper-wasm-v1/voices/ATTRIBUTION.txt`.
+The Piper engine, ONNX WebAssembly runtime, and all three Swedish medium voice models (Alma, Lisa, and NST) are served by the separate `ai-spel-static` Nginx service from same-origin `/static/piper-wasm-v1/`. Its own Gitea pipeline downloads pinned models, verifies sizes and hashes, and publishes a versioned image and Helm chart independently from frontend changes. The frontend's Piper adapter redirects the library's built-in model URLs to this static service; it does not request voice files from Hugging Face. Voice models are fetched on first use and cached in browser storage when supported. Dialogue text is synthesized on the device and is not sent to a TTS provider. Attribution is available at `/static/piper-wasm-v1/voices/ATTRIBUTION.txt`.
 
 The microphone button beside the text field uses Web Speech API when available and sends only its final Swedish transcript through the same submission path as typed text. Before its first use, the client explains that the browser may process audio through a provider-selected service; the game itself does not receive or store recordings, while the final transcript is saved as ordinary dialogue. Writing `röstinställningar` or saying `röstkommando alternativ` opens temporary voice controls in the dialogue. The player can adjust speaking rate and enable or disable short audio cues. Voice preferences are stored in browser `localStorage`.
 
@@ -24,7 +26,9 @@ System voice commands require the prefix `röstkommando`: `upprepa`, `avbryt`, `
 
 On iPhone and iPad, Safari requires a user gesture to unlock Web Audio. The frontend resumes the audio context during a trusted tap or key press and then reads generated replies automatically. Voice settings include **Testa ljudet**; if a game reply does not start, a **Läs senaste repliken** button appears in the dialogue. First use can take longer while the voice model downloads; progress and errors are announced in the dialogue. Actual iPhone/iPad playback remains a release acceptance check and must not be claimed until tested on a device.
 
-The browser and API stay on the same public origin, `https://ai-spel.k3s.k0.nu`. The ingress routes `/` to this frontend service and routes `/api`, `/version`, and `/health` to the Rust service. That keeps the current server-side session cookie same-origin and avoids browser CORS configuration.
+The browser and API stay on the same public origin, `https://ai-spel.k3s.k0.nu`. The ingress routes `/` to this frontend service and routes `/api`, `/graphql`, `/version`, and `/health` to the Rust service. Player dialogue is read through the authenticated, cursor-paginated GraphQL `dialogue(first, before)` query. Admin conversation traces use the audited `adminPlayerTrace(accountId, limit, offset)` GraphQL query. Writes remain explicit Rust command routes. That keeps the current server-side session cookie same-origin and avoids browser CORS configuration.
+
+The public footer displays the frontend, static-service, and API versions. It fetches `/static/version.json` without caching; the static service creates that file from its sole `VERSION` source during the same image build used for the matching Helm chart and release tag.
 
 ## Release flow
 
@@ -35,3 +39,6 @@ CI needs repository Action secrets named `HARBOR_USERNAME` and `HARBOR_PASSWORD`
 The first Argo Application is described in `deploy/argocd/application.yaml`. The backend ingress routes `/api`, `/version`, and health checks to Rust while serving the static client at `/`. Both services share the same origin so the session cookie remains same-origin.
 
 Builds, checks, and runtime verification happen in the pipeline and Kubernetes. Do not run the app, tests, container images, or binaries locally.
+# Browser-local AI status
+
+Gemma/Wllama assistance is currently paused. `FEATURE_FLAGS.localAI` in `index.html` is `false`; while disabled, the client does not load the adapter, download model/runtime assets, run local inference, or send local hints. Re-enabling it requires an explicit code change and the normal frontend release pipeline. This does not affect the regular Rust-backed dialogue flow.

VERSION

@@ -1 +1 @@
-0.1.47
+0.1.74

deploy/helm/ai-spel-frontend/Chart.yaml

@@ -2,5 +2,5 @@ apiVersion: v2
 name: ai-spel-frontend
 description: Static web frontend for the Swedish life simulator
 type: application
-version: 0.1.47
+version: 0.1.74
-appVersion: "0.1.47"
+appVersion: "0.1.74"

docs/adr/0001-piper-es-module-worker.md

@@ -0,0 +1,13 @@
+# ADR 0001: Run Piper in an ES module worker
+
+## Decision
+
+Bundle the Piper worker as an ES module and start it with `new Worker(url, { type: 'module' })`.
+
+## Context
+
+Piper loads ONNX Runtime Web and its WebAssembly loader dynamically. The current IIFE/classic worker bundle reports `TypeError: "undefined" cannot be parsed as a URL` during WASM backend initialization, even though the pinned same-origin ONNX loader and binary are available. This is consistent with module-relative URL metadata being lost in the classic bundle.
+
+## Consequences
+
+The browser must support module workers. Current supported browser targets already require modern WebAssembly and worker support. CI asserts that the worker is built and started in module mode; production acceptance remains the real deployed browser flow.

docs/adr/0002-single-thread-onnx-for-mobile-tts.md

@@ -0,0 +1,13 @@
+# ADR 0002: Run ONNX with one thread inside the Piper worker
+
+## Decision
+
+Configure ONNX Runtime Web to use one thread and stop backend initialization after 60 seconds. Piper already runs in a dedicated worker, so ONNX does not need to create another worker pool.
+
+## Context
+
+On iPad, the threaded ONNX WASM file downloaded fully, but the runtime did not report that initialization had completed. Nested worker startup and shared-memory setup are a likely compatibility boundary on mobile Safari. The available production assets currently include the threaded ONNX build.
+
+## Consequences
+
+This removes ONNX's additional worker pool and bounds initialization failures with a visible error. It may reduce inference throughput on desktop devices. CI and Kubernetes release verification must confirm that Piper still initializes and speaks on supported browsers; iPad acceptance remains a manual device check.

docs/adr/0003-limit-piper-wasm-to-one-thread.md

@@ -0,0 +1,13 @@
+# ADR 0003: Limit Piper WASM to one thread and reset failed workers
+
+## Decision
+
+The Piper worker reports one available processor to Piper and configures ONNX Runtime for one thread. If a synthesis request fails, the browser client terminates that worker; the next player retry creates a fresh worker and ONNX module.
+
+## Context
+
+Piper TTS Web 1.0.3 sets `env.wasm.numThreads` from `navigator.hardwareConcurrency` during `TtsSession.init()`, after the application had configured ONNX Runtime. On iPad this could undo the single-thread setting and request a larger WASM worker pool. The library also retains a singleton session if ONNX initialization fails, so retrying in the same worker can reuse a failed initialization.
+
+## Consequences
+
+The one-thread limit reduces parallel inference but lowers memory pressure on mobile Safari. Failed initialization no longer poisons future attempts in the same worker. This cannot guarantee that every device has enough memory for the 63 MB medium voice model; the CI release and Kubernetes checks validate packaging and deployment, while actual iPad memory behavior still requires device verification.

docs/adr/0004-pause-browser-local-ai.md

@@ -0,0 +1,19 @@
+# ADR 0004: Pause browser-local AI by default
+
+## Status
+
+Accepted
+
+## Decision
+
+Browser-local Gemma/Wllama assistance is disabled by the `FEATURE_FLAGS.localAI` flag in `index.html`. The adapter is loaded dynamically only when the flag is true. When it is false, the browser does not load the adapter, request model/runtime assets, initialize local inference, attach local hints to dialogue requests, or produce local offline replies. The normal text dialogue remains available through the Rust API.
+
+## Rationale
+
+Pause model execution while the rest of the game and narrator experience is stabilized. Keep the implementation available for a deliberate future reactivation without downloading model resources or sending local inference hints in the meantime.
+
+## Consequences
+
+- Re-enabling local AI requires an explicit code change and frontend release.
+- Tests and the static frontend check must enforce the default-off behavior and lazy loading.
+- Existing Gemma asset publication and documentation may remain in place; the browser does not request those assets while the flag is off.

nginx/default.conf

@@ -12,9 +12,10 @@ server {
     location ^~ /tts/piper-wasm-v1/ {
         types {
             application/wasm wasm;
-            application/javascript js;
+            application/javascript js mjs;
             application/json json;
             application/octet-stream onnx;
+            text/plain txt;
         }
         add_header Cache-Control "public, max-age=31536000, immutable" always;
         add_header Cross-Origin-Opener-Policy "same-origin" always;

piper-download-progress.mjs

@@ -0,0 +1,37 @@
+export function trackDownloadProgress(response, {id, postMessage, asset = 'model', file}) {
+  const reader = response.body?.getReader?.();
+  if (!reader || typeof ReadableStream !== 'function' || typeof Response !== 'function') return response;
+
+  const total = Number(response.headers.get('content-length')) || 0;
+  let loaded = 0;
+  const body = new ReadableStream({
+    async pull(controller) {
+      try {
+        const {done, value} = await reader.read();
+        if (done) {
+          if (asset === 'runtime') {
+            postMessage({id, type: 'progress', progress: {stage: 'initializing-runtime', file}});
+          }
+          controller.close();
+          return;
+        }
+        loaded += value.byteLength;
+        const progress = {asset, loaded, total};
+        if (file) progress.file = file;
+        postMessage({id, type: 'progress', progress});
+        controller.enqueue(value);
+      } catch (error) {
+        controller.error(error);
+      }
+    },
+    cancel(reason) {
+      return reader.cancel(reason);
+    },
+  });
+
+  return new Response(body, {
+    status: response.status,
+    statusText: response.statusText,
+    headers: response.headers,
+  });
+}

piper-model-routing.mjs

@@ -1,5 +1,5 @@
 const PIPER_MODEL_ORIGIN = 'https://huggingface.co/diffusionstudio/piper-voices/resolve/main/';
-const localPathPrefix = '/tts/piper-wasm-v1/voices/';
+const localPathPrefix = '/static/piper-wasm-v1/voices/';
 
 export function localPiperModelUrl(requestedUrl, origin) {
   if (!requestedUrl.startsWith(PIPER_MODEL_ORIGIN)) return null;

piper-onnx-config.mjs

@@ -0,0 +1,31 @@
+const assetRoot = '/static/piper-wasm-v1/onnx/';
+
+export function createOnnxWasmPaths(origin) {
+  const absoluteAssetRoot = new URL(assetRoot, origin).href;
+  return absoluteAssetRoot;
+}
+
+export function configureOnnxRuntime(onnxRuntime) {
+  if (!onnxRuntime?.env?.wasm) throw new TypeError('ONNX Runtime WebAssembly settings are unavailable.');
+  // Piper 1.0.3 overwrites this value in TtsSession.init() with
+  // navigator.hardwareConcurrency. Override that source as well, otherwise an
+  // iPad may start a large WASM worker pool despite this setting.
+  const navigatorObject = globalThis.navigator;
+  if (navigatorObject && navigatorObject.hardwareConcurrency !== 1) {
+    try {
+      Object.defineProperty(navigatorObject, 'hardwareConcurrency', {
+        configurable: true,
+        value: 1,
+      });
+    } catch (_) {
+      // The check below converts an unsupported override into a clear error.
+    }
+  }
+  if (navigatorObject && navigatorObject.hardwareConcurrency !== 1) {
+    throw new Error('Piper could not be restricted to one processor thread.');
+  }
+  // Piper already runs in a worker. Avoid a second worker pool on mobile Safari.
+  onnxRuntime.env.wasm.numThreads = 1;
+  onnxRuntime.env.wasm.initTimeout = 60000;
+}
+

piper-runtime.js

@@ -6,7 +6,9 @@ const pending = new Map();
 
 function getWorker() {
   if (worker) return worker;
-  worker = new Worker(workerUrl);
+  // ONNX Runtime resolves its WASM loader relative to the worker module URL.
+  // Keep the worker as ESM so import.meta.url remains available at runtime.
+  worker = new Worker(workerUrl, {type: 'module'});
   worker.onmessage = event => {
     const message = event.data || {};
     const request = pending.get(message.id);
@@ -15,8 +17,19 @@ function getWorker() {
       request.onProgress?.(message.progress);
     } else {
       pending.delete(message.id);
-      if (message.type === 'audio') request.resolve(new Blob([message.audio], {type: 'audio/x-wav'}));
+      if (message.type === 'audio') {
-      else request.reject(new Error(message.error || 'Piper-arbetaren misslyckades.'));
+        request.resolve(new Blob([message.audio], {type: 'audio/x-wav'}));
+      } else {
+        const error = new Error(message.error || 'Piper-arbetaren misslyckades.');
+        request.reject(error);
+        // TtsSession keeps a singleton even when ONNX initialization fails.
+        // Discard the whole worker so a player retry starts with a clean ONNX
+        // module and does not reuse a poisoned WASM initialization.
+        for (const queued of pending.values()) queued.reject(error);
+        pending.clear();
+        worker?.terminate();
+        worker = null;
+      }
     }
   };
   worker.onerror = event => {

piper-worker.js

@@ -1,30 +1,63 @@
 import {TtsSession} from '@mintplex-labs/piper-tts-web';
+import * as onnxRuntime from 'onnxruntime-web';
+import {trackDownloadProgress} from './piper-download-progress.mjs';
 import {localPiperModelUrl} from './piper-model-routing.mjs';
+import {configureOnnxRuntime, createOnnxWasmPaths} from './piper-onnx-config.mjs';
 
 const VOICE_ID = 'sv_SE-nst-medium';
-const ASSET_ROOT = '/tts/piper-wasm-v1';
+const ASSET_ROOT = '/static/piper-wasm-v1';
 const localFetch = self.fetch.bind(self);
 
-// The pinned TTS library uses a fixed Hugging Face URL. Keep its OPFS cache logic,
+function reportStage(id, stage, details = {}) {
-// but serve every model/config request from this frontend origin instead.
+  if (Number.isSafeInteger(id)) self.postMessage({id, type: 'progress', progress: {stage, ...details}});
-self.fetch = (resource, options) => {
+}
+
+// Keep the TTS library's OPFS cache logic, but serve models/config from the
+// same-origin static asset service instead of its built-in external model URL.
+self.fetch = async (resource, options) => {
   const requestedUrl = resource instanceof Request ? resource.url : String(resource);
   const localModelUrl = localPiperModelUrl(requestedUrl, self.location.origin);
-  if (!localModelUrl) return localFetch(resource, options);
+  if (!localModelUrl) {
+    const requestUrl = new URL(requestedUrl, self.location.origin);
+    if (requestUrl.origin !== self.location.origin) throw new Error(`External Piper asset blocked: ${requestUrl.origin}`);
+    const runtimeAsset = requestedUrl.includes('/static/piper-wasm-v1/onnx/');
+    const runtimeFile = runtimeAsset ? new URL(requestedUrl, self.location.origin).pathname.split('/').pop() : undefined;
+    if (activeRequestId && runtimeAsset) {
+      reportStage(activeRequestId, 'loading-runtime-assets', {file: runtimeFile});
+    }
+    const response = await localFetch(resource, options);
+    if (!activeRequestId || !runtimeAsset) return response;
+    reportStage(activeRequestId, 'reading-runtime-assets', {file: runtimeFile});
+    return trackDownloadProgress(response, {
+      id: activeRequestId,
+      asset: 'runtime',
+      file: runtimeFile,
+      postMessage: message => self.postMessage(message),
+    });
+  }
+  const isModel = localModelUrl.endsWith('.onnx');
+  if (activeRequestId) reportStage(activeRequestId, isModel ? 'loading-model' : 'loading-model-info');
   const redirected = resource instanceof Request ? new Request(localModelUrl, resource) : localModelUrl;
-  return localFetch(redirected, options);
+  const response = await localFetch(redirected, options);
+  if (!activeRequestId || !isModel) return response;
+  reportStage(activeRequestId, 'reading-model');
+  return trackDownloadProgress(response, {
+    id: activeRequestId,
+    postMessage: message => self.postMessage(message),
+  });
 };
 
 let sessionPromise;
 let inferenceQueue = Promise.resolve();
+let activeRequestId = null;
 
-async function getSession(onProgress) {
+async function getSession() {
   if (!sessionPromise) {
+    configureOnnxRuntime(onnxRuntime);
     sessionPromise = TtsSession.create({
       voiceId: VOICE_ID,
-      progress: onProgress,
       wasmPaths: {
-        onnxWasm: `${ASSET_ROOT}/onnx/`,
+        onnxWasm: createOnnxWasmPaths(self.location.origin),
         piperData: `${ASSET_ROOT}/piper/piper_phonemize.data`,
         piperWasm: `${ASSET_ROOT}/piper/piper_phonemize.wasm`,
       },
@@ -39,13 +72,18 @@ async function getSession(onProgress) {
 async function synthesizeMessage(event) {
   const {id, text} = event.data || {};
   if (!Number.isSafeInteger(id) || typeof text !== 'string') return;
+  activeRequestId = id;
+  reportStage(id, 'loading-runtime');
   try {
-    const session = await getSession(progress => self.postMessage({id, type: 'progress', progress}));
+    const session = await getSession();
+    reportStage(id, 'model-ready');
     const blob = await session.predict(text);
     const audio = await blob.arrayBuffer();
     self.postMessage({id, type: 'audio', audio}, [audio]);
   } catch (error) {
     self.postMessage({id, type: 'error', error: String(error?.message || error)});
+  } finally {
+    activeRequestId = null;
   }
 }
 

scripts/build-piper.mjs

@@ -1,80 +1,13 @@
 import {build} from 'esbuild';
-import {createHash} from 'node:crypto';
+import {mkdir, rm, readFile} from 'node:fs/promises';
-import {createWriteStream} from 'node:fs';
-import {cp, mkdir, readdir, rm, stat, rename, unlink} from 'node:fs/promises';
 import {join, resolve} from 'node:path';
-import {Readable, Transform} from 'node:stream';
-import {pipeline} from 'node:stream/promises';
 
 const root = process.cwd();
 const publicRoot = join(root, 'public');
-const assetRoot = join(publicRoot, 'tts', 'piper-wasm-v1');
+const workerRoot = join(publicRoot, 'tts', 'piper-wasm-v1');
-const onnxSource = join(root, 'node_modules', 'onnxruntime-web', 'dist');
-const piperSource = join(root, 'node_modules', '@diffusionstudio', 'piper-wasm', 'build');
-const piperWasm = join(piperSource, 'piper_phonemize.wasm');
-const piperData = join(piperSource, 'piper_phonemize.data');
 
-const voiceRevision = 'c10ece1aade47bb51c153c893d14e5bf8e5b7117';
+await rm(workerRoot, {recursive: true, force: true});
-const voiceRepository = 'https://huggingface.co/rhasspy/piper-voices/resolve';
+await mkdir(workerRoot, {recursive: true});
-const voiceFiles = [
-  {path: 'sv/sv_SE/alma/medium/sv_SE-alma-medium.onnx', size: 63434611, sha256: 'f75455eb6afeb1db9e2905d5bd7e4115afcd3710e0d399b6a40981766d32fedc'},
-  {path: 'sv/sv_SE/alma/medium/sv_SE-alma-medium.onnx.json', size: 6059, gitBlobSha1: '7eaf3ef20af4a74e9cee9d29b98de3d9fcec0bdc'},
-  {path: 'sv/sv_SE/lisa/medium/sv_SE-lisa-medium.onnx', size: 63511038, sha256: '94cae912b31d6e9140d3f5160f1815951588600c7a9e43d539ba1e81a110d131'},
-  {path: 'sv/sv_SE/lisa/medium/sv_SE-lisa-medium.onnx.json', size: 7239, gitBlobSha1: 'b2b5fb0f8af2aae4ddfb81eaa7709c6b8faa412b'},
-  {path: 'sv/sv_SE/nst/medium/sv_SE-nst-medium.onnx', size: 63104526, sha256: 'df011f56825a59dd1efc080c38a65a1ef70407e60f63050e9246f43a3d7e471e'},
-  {path: 'sv/sv_SE/nst/medium/sv_SE-nst-medium.onnx.json', size: 4157, gitBlobSha1: '9766b9769436046a610a505fda925032e12a66b3'},
-];
-
-async function downloadVoiceFile(file) {
-  const destination = join(assetRoot, 'voices', file.path);
-  const temporary = `${destination}.part`;
-  await mkdir(join(destination, '..'), {recursive: true});
-  const response = await fetch(`${voiceRepository}/${voiceRevision}/${file.path}`);
-  if (!response.ok || !response.body) {
-    throw new Error(`Unable to download Piper voice asset ${file.path}: HTTP ${response.status}`);
-  }
-  const sha256 = createHash('sha256');
-  const sha1 = createHash('sha1').update(Buffer.from(`blob ${file.size}\0`));
-  let size = 0;
-  const verifyAndPassThrough = new Transform({
-    transform(chunk, _encoding, callback) {
-      size += chunk.length;
-      sha256.update(chunk);
-      sha1.update(chunk);
-      callback(null, chunk);
-    },
-  });
-  try {
-    await pipeline(Readable.fromWeb(response.body), verifyAndPassThrough, createWriteStream(temporary));
-    const actualSha1 = sha1.digest('hex');
-    if (size !== file.size || (file.sha256 && sha256.digest('hex') !== file.sha256) ||
-        (file.gitBlobSha1 && actualSha1 !== file.gitBlobSha1)) {
-      throw new Error(`Piper voice asset integrity check failed for ${file.path} (${size} bytes)`);
-    }
-    await rename(temporary, destination);
-  } catch (error) {
-    await unlink(temporary).catch(() => {});
-    throw error;
-  }
-}
-
-await rm(assetRoot, {recursive: true, force: true});
-await mkdir(join(assetRoot, 'onnx'), {recursive: true});
-await mkdir(join(assetRoot, 'piper'), {recursive: true});
-for (const file of voiceFiles) await downloadVoiceFile(file);
-await cp(join(root, 'PIPER_VOICE_NOTICES.txt'), join(assetRoot, 'voices', 'ATTRIBUTION.txt'));
-
-const wasmFiles = (await readdir(onnxSource))
-  .filter(name => name.endsWith('.wasm'));
-if (wasmFiles.length === 0) throw new Error('ONNX Runtime WebAssembly files are missing');
-for (const file of wasmFiles) {
-  await cp(join(onnxSource, file), join(assetRoot, 'onnx', file));
-}
-for (const file of [piperWasm, piperData]) {
-  const fileStat = await stat(file).catch(() => null);
-  if (!fileStat?.isFile() || fileStat.size === 0) throw new Error(`Missing Piper WebAssembly asset: ${file}`);
-  await cp(file, join(assetRoot, 'piper', file.split('/').at(-1)));
-}
 
 await build({
   entryPoints: [resolve(root, 'piper-runtime.js')],
@@ -90,9 +23,9 @@ await build({
 
 await build({
   entryPoints: [resolve(root, 'piper-worker.js')],
-  outfile: join(assetRoot, 'piper-worker.js'),
+  outfile: join(workerRoot, 'piper-worker.js'),
   bundle: true,
-  format: 'iife',
+  format: 'esm',
   platform: 'browser',
   target: ['es2022'],
   external: ['fs', 'path'],
@@ -100,14 +33,14 @@ await build({
   legalComments: 'none',
 });
 
-for (const file of voiceFiles) {
+const workerBundle = await readFile(join(workerRoot, 'piper-worker.js'), 'utf8');
-  const downloaded = join(assetRoot, 'voices', file.path);
+for (const route of [
-  if ((await stat(downloaded)).size !== file.size) throw new Error(`Missing Piper voice asset: ${downloaded}`);
+  'https://huggingface.co/diffusionstudio/piper-voices/resolve/main/',
-}
+  '/static/piper-wasm-v1',
-const workerBundle = await (await import('node:fs/promises')).readFile(join(assetRoot, 'piper-worker.js'), 'utf8');
+  '/static/piper-wasm-v1/voices/',
-if (!workerBundle.includes('https://huggingface.co/diffusionstudio/piper-voices/resolve/main/') ||
+  '/static/piper-wasm-v1/onnx/',
-    !workerBundle.includes('/tts/piper-wasm-v1/voices/')) {
+]) {
-  throw new Error('The Piper worker is missing its same-origin model download route.');
+  if (!workerBundle.includes(route)) throw new Error(`The Piper worker is missing static route ${route}`);
 }
 
-console.log(`Piper browser assets prepared at ${assetRoot}`);
+console.log(`Piper worker bundle prepared at ${workerRoot}`);

scripts/check-frontend.sh

@@ -5,14 +5,21 @@ test -s index.html
 grep -q '<html lang="sv">' index.html
 grep -q 'id="frontend-version">__FRONTEND_VERSION__' index.html
 grep -q '<script src="/voice.js?v=__FRONTEND_VERSION__"></script>' index.html
+grep -q 'const FEATURE_FLAGS = Object.freeze({localAI: false})' index.html
+grep -q 'if (!FEATURE_FLAGS.localAI) return Promise.resolve(null)' index.html
+grep -q "script.src = '/static/' + 'wllama-gemma3-v2/local-ai.js'" index.html
+! grep -q '<script src="/static/wllama-gemma3-v2/local-ai.js"></script>' index.html
 grep -q '<script src="/piper-runtime.js?v=__FRONTEND_VERSION__"></script>' index.html
 grep -q 'no-store, no-cache, must-revalidate, proxy-revalidate, max-age=0' nginx/default.conf
 grep -q 'Cross-Origin-Embedder-Policy "require-corp"' nginx/default.conf
 grep -q 'add_header Cross-Origin-Embedder-Policy "require-corp" always;' nginx/default.conf
 grep -q 'add_header Cross-Origin-Resource-Policy "same-origin" always;' nginx/default.conf
-grep -q 'application/javascript js;' nginx/default.conf
+grep -q 'application/javascript js mjs;' nginx/default.conf
 grep -q "request('/version')" index.html
+grep -q 'id="static-version">hämtar version…' index.html
+grep -q "request('/static/version.json', {cache: 'no-store'})" index.html
 grep -Fq "credentials: 'same-origin'" index.html
+grep -q "request('/graphql'" index.html
 grep -q 'id="dialogue" aria-label="Dialoghistorik" role="log" aria-live="polite"' index.html
 grep -q 'id="message" name="message"' index.html
 grep -q 'id="auth-intro"' index.html
@@ -26,7 +33,13 @@ grep -q "retry.textContent = 'Läs senaste repliken'" index.html
 grep -q "location.pathname !== '/admin'" index.html
 grep -q "if (options.speak !== false && location.pathname !== '/admin')" index.html
 grep -q 'id="microphone" aria-pressed="false"' index.html
+grep -q 'type="submit" id="send-message">Skicka</button>' index.html
 grep -q 'id="voice-status" role="status" aria-live="polite"' index.html
+grep -q 'id="speech-progress" aria-label="Nedladdning av svensk röstmodell"' index.html
+grep -q 'id="speech-progress-value"' index.html
+grep -q 'MB hämtade' index.html
+grep -q 'loading-runtime-assets' piper-worker.js
+grep -q 'Röstmodellen är inläst' voice.js
 grep -q "window.GameVoice.createVoiceController" index.html
 grep -q "addVoiceButton(item, 'Testa ljudet'" index.html
 grep -q 'Jag förstår – starta mikrofonen' index.html
@@ -38,18 +51,34 @@ grep -q 'WebKit versions before Safari 27' voice.js
 grep -q "voiceIdentity: speakingNpc.voice_identity" index.html
 grep -q "role: 'narrator', voiceIdentity: 'neutral'" index.html
 grep -q 'COPY voice.js /usr/share/nginx/html/voice.js' Dockerfile
+grep -Fq 'if (!currentAccount.is_admin && FEATURE_FLAGS.localAI)' index.html
+grep -Fq 'localAI?.start(currentAccount.id)' index.html
+grep -q "type === 'terms-required'" index.html
+grep -q 'Gemma-villkor' index.html
+grep -q 'lokal ai godkänn villkor' index.html
+grep -q 'lokal ai avvisa villkor' index.html
+grep -q 'client_assistance: clientAssistance' index.html
 grep -q 'COPY public/piper-runtime.js /usr/share/nginx/html/piper-runtime.js' Dockerfile
 grep -q 'COPY public/tts/piper-wasm-v1/' Dockerfile
 grep -q 'build:piper' .gitea/workflows/publish.yaml
 grep -q 'Piper' piper-runtime.js
 grep -q 'sv_SE-nst-medium' piper-worker.js
-grep -q 'new Worker(workerUrl)' piper-runtime.js
+grep -q "new Worker(workerUrl, {type: 'module'})" piper-runtime.js
 grep -q 'worker.js?v=' piper-runtime.js
 grep -q 'worker.onmessageerror' piper-runtime.js
 grep -q 'localPiperModelUrl' piper-worker.js
+grep -q 'External Piper asset blocked' piper-worker.js
+grep -q 'trackDownloadProgress' piper-worker.js
+grep -q 'new ReadableStream' piper-download-progress.mjs
+grep -q 'createOnnxWasmPaths(self.location.origin)' piper-worker.js
+grep -q 'configureOnnxRuntime(onnxRuntime)' piper-worker.js
+grep -q 'numThreads = 1' piper-onnx-config.mjs
+grep -q "hardwareConcurrency'" piper-onnx-config.mjs
+grep -q 'Piper could not be restricted to one processor thread' piper-onnx-config.mjs
+grep -q '/static/piper-wasm-v1/onnx/' piper-onnx-config.mjs
 grep -q 'PIPER_MODEL_ORIGIN' piper-model-routing.mjs
-grep -q '/tts/piper-wasm-v1/voices/' piper-model-routing.mjs
+grep -q '/static/piper-wasm-v1/voices/' piper-model-routing.mjs
-grep -q 'same-origin.*voice' README.md
+grep -q 'ai-spel-static' README.md
 grep -q 'ATTRIBUTION.txt' index.html
 grep -q "SpeechRecognition || win.webkitSpeechRecognition" voice.js
 grep -q "submitMessage(text, 'voice')" index.html
@@ -65,13 +94,38 @@ grep -q "Säg till Robin" index.html
 grep -q "result.intent === 'move' || result.intent === 'meet'" index.html
 grep -q 'dialogue_temporarily_unavailable:' index.html
 grep -q 'id="admin-view"' index.html
+grep -q 'id="play-layout"' index.html
+grep -q 'id="debug-panel"' index.html
+grep -q "query AdminPlayerTrace" index.html
+grep -q 'adminPlayerTrace(accountId: \$accountId, limit: \$limit, offset: \$offset)' index.html
+grep -q 'function renderDecisionTrace' index.html
+grep -q 'appendFormattedJson(pre, group)' index.html
+grep -Fq "JSON.stringify(value, null, '\\t')" index.html
+grep -q 'json-token-key' index.html
+grep -q 'json-token-string' index.html
+grep -q 'debug-enabled' index.html
 grep -q '/api/admin/players?' index.html
+grep -q 'Välj konto att följa' index.html
+grep -q 'function renderAdminConversations' index.html
+grep -q 'function renderDecisionProvenance' index.html
+grep -q 'Visa äldre samtalsturer' index.html
+grep -q 'Lättläst konversation' index.html
+grep -q 'admin-simple-conversation' index.html
+grep -q 'aria-expanded' index.html
 grep -q "request('/api/admin/prompts'" index.html
 grep -q "request('/api/admin/rules'" index.html
 grep -q "request('/api/admin/settings/logging'" index.html
 grep -q 'data-admin-view="logging"' index.html
 grep -q '/api/admin/audit?limit=200' index.html
-grep -q 'entry.data.external_call' index.html
+grep -q 'data-admin-view="npcs"' index.html
+grep -q '/api/admin/npcs?q=' index.html
+grep -q '/api/admin/npcs/${encodeURIComponent(npcId)}/profile' index.html
+grep -q '/api/admin/npcs/${encodeURIComponent(npcId)}/identity' index.html
+grep -q '/api/admin/npcs/${encodeURIComponent(npcId)}/memories/${encodeURIComponent(memory.id)}' index.html
+grep -q '/api/admin/npcs/${encodeURIComponent(npcId)}/possessions' index.html
+grep -q 'Egenskaper och framtida profilfält' index.html
+grep -q 'Minnesrekonstruktioner' index.html
+grep -q 'interaction.external_call' index.html
 grep -q 'Externt tjänsteanrop' index.html
 grep -q "request('/api/world/start'" index.html
 grep -q "request('/api/character'" index.html
@@ -80,8 +134,8 @@ if grep -Eq 'id="(new-character|world-setup|shop|buy-offer|balance|inventory|pla
   echo 'The active player UI must not contain permanent data panels, forms, or action buttons.' >&2
   exit 1
 fi
-if sed -n '/<section id="play"/,/^    <\/section>/p' index.html | grep '<button' | grep -v 'id="microphone"' >/dev/null; then
+if sed -n '/<section id="play"/,/^    <\/section>/p' index.html | grep '<button' | grep -Ev 'id="(microphone|send-message)"' >/dev/null; then
-  echo 'Only the microphone control may appear beside the game text input.' >&2
+  echo 'Only the send and microphone controls may appear beside the game text input.' >&2
   exit 1
 fi
 

tests/activity-feedback.test.cjs

@@ -0,0 +1,23 @@
+const assert = require('node:assert/strict');
+const fs = require('node:fs');
+const path = require('node:path');
+const test = require('node:test');
+
+const source = fs.readFileSync(path.join(__dirname, '..', 'index.html'), 'utf8');
+
+test('player sees an accessible live activity line during world and character proposals', () => {
+  assert.match(source, /function setActivityStatus\(message\)/);
+  assert.match(source, /setAttribute\('aria-live', 'polite'\)/);
+  assert.match(source, /AI ta fram ett förslag på startvärld och en plats/);
+  assert.match(source, /AI skapa eller uppdatera ett biografiförslag/);
+});
+
+test('dialogue processing explains that scene, place, or person establishment is being considered', () => {
+  assert.match(source, /Berättaren arbetar med din avsikt\. Den avgör om en plats, scen eller person behöver etableras/);
+});
+
+test('local AI activity is shown separately from general narrator processing', () => {
+  assert.match(source, /detail\.type === 'activity' \? 'data-local-ai-activity'/);
+  assert.match(source, /Lokal AI: \$\{detail\.text \|\| ''\}/);
+  assert.match(source, /clearActivityStatus\(\);/);
+});

tests/admin-conversation-trace.test.cjs

@@ -0,0 +1,57 @@
+const test = require('node:test');
+const assert = require('node:assert/strict');
+const fs = require('node:fs');
+
+const page = fs.readFileSync(new URL('../index.html', `file://${__filename}`), 'utf8');
+
+test('admin can select a searched account before opening its conversations', () => {
+  assert.match(page, /Välj konto att följa/);
+  assert.match(page, /chooserSelect\.value\);/);
+  assert.match(page, /\/api\/admin\/players\?q=/);
+});
+
+test('conversation transcript exposes player and narrator messages as expandable replies', () => {
+  assert.match(page, /function renderAdminConversations/);
+  assert.match(page, /payload\.player_message/);
+  assert.match(page, /payload\.reply \|\| payload\.npc_reply/);
+  assert.match(page, /aria-expanded/);
+  assert.match(page, /renderDecisionProvenance\(group, event, detail\)/);
+});
+
+test('selected replies reveal AI, Luau and Rust provenance without injecting HTML', () => {
+  assert.match(page, /function renderDecisionProvenance/);
+  assert.match(page, /interaction\.prompt_yaml/);
+  assert.match(page, /interaction\.external_call/);
+  assert.match(page, /details\.luau_trace/);
+  assert.match(page, /details\.rust_validation/);
+  assert.match(page, /element\.textContent = text/);
+});
+
+test('account trace can be paged until all stored turns are loaded', () => {
+  assert.match(page, /Visa äldre samtalsturer/);
+  assert.match(page, /query AdminPlayerTrace\(\$accountId: String, \$limit: Int!, \$offset: Int!\)/);
+  assert.match(page, /getAdminTrace\(accountId, trace\.limit \|\| 200, adminTraceOffset\)/);
+});
+
+test('admin can switch to a storyteller-only reading view and return to diagnostics', () => {
+  assert.match(page, /Lättläst konversation/);
+  assert.match(page, /Visa felsökningsspår/);
+  assert.match(page, /root\.classList\.toggle\('admin-simple-conversation', adminSimpleConversation\)/);
+  assert.match(page, /playerLine\.className = 'admin-story-player'/);
+  assert.match(page, /story\.className = 'admin-story-reply'/);
+  assert.match(page, /article > :not\(\.admin-story-player\):not\(\.admin-story-reply\)/);
+  assert.match(page, /article > \.admin-story-player,\s*#admin-content \.admin-simple-conversation article > \.admin-story-reply/);
+});
+
+
+test('admin JSON responses use tab-indented syntax-colored accessible output', () => {
+  assert.match(page, /function appendFormattedJson/);
+  assert.match(page, /JSON\.stringify\(value, null, '\\t'\)/);
+  assert.match(page, /json-token-key/);
+  assert.match(page, /json-token-string/);
+  assert.match(page, /json-token-number/);
+  assert.match(page, /json-token-boolean/);
+  assert.match(page, /json-token-null/);
+  assert.match(page, /setAttribute\('aria-label', 'Formaterat JSON-svar'\)/);
+  assert.match(page, /appendFormattedJson\(pre, group\)/);
+});

tests/local-ai-integration.test.cjs

@@ -0,0 +1,31 @@
+const assert = require('node:assert/strict');
+const fs = require('node:fs');
+const path = require('node:path');
+const test = require('node:test');
+
+const page = fs.readFileSync(path.join(__dirname, '..', 'index.html'), 'utf8');
+
+test('local AI is paused by default and adapter loading is gated by its feature flag', () => {
+  assert.match(page, /const FEATURE_FLAGS = Object\.freeze\(\{localAI: false\}\)/);
+  assert.match(page, /if \(!FEATURE_FLAGS\.localAI\) return Promise\.resolve\(null\)/);
+  assert.match(page, /script\.src = '\/static\/' \+ 'wllama-gemma3-v2\/local-ai\.js'/);
+  assert.doesNotMatch(page, /<script src="\/static\/wllama-gemma3-v2\/local-ai\.js"><\/script>/);
+  assert.match(page, /if \(!currentAccount\.is_admin && FEATURE_FLAGS\.localAI\)[\s\S]*?loadLocalAiAdapter\(\)/);
+});
+
+test('Gemma license links are shown in the accessible game dialogue and consent can be withdrawn', () => {
+  assert.match(page, /type === 'terms-required'/);
+  assert.match(page, /Gemma-villkor/);
+  assert.match(page, /Spelvillkor för Gemma/);
+  assert.match(page, /policy för förbjuden användning/);
+  assert.match(page, /lokal ai godkänn villkor/);
+  assert.match(page, /lokal ai avvisa villkor/);
+  assert.match(page, /window\.localGameAI\?\.stop\(\)/);
+});
+
+test('local assistance remains advisory and preserves the normal dialogue request path', () => {
+  assert.match(page, /client_assistance: clientAssistance/);
+  assert.match(page, /Local assistance is optional; the original input always proceeds/);
+  assert.match(page, /request\('\/api\/dialogue'/);
+  assert.match(page, /Lokalt, osparat förslag:/);
+});

tests/piper-download-progress.test.cjs

@@ -0,0 +1,59 @@
+const assert = require('node:assert/strict');
+const test = require('node:test');
+
+test('model response progress forwards each downloaded chunk and preserves the response', async () => {
+  const {trackDownloadProgress} = await import('../piper-download-progress.mjs');
+  const source = new ReadableStream({
+    start(controller) {
+      controller.enqueue(new Uint8Array([1, 2]));
+      controller.enqueue(new Uint8Array([3, 4, 5]));
+      controller.close();
+    },
+  });
+  const response = new Response(source, {headers: {'content-length': '5'}});
+  const messages = [];
+  const tracked = trackDownloadProgress(response, {id: 7, postMessage: message => messages.push(message)});
+
+  assert.deepEqual([...new Uint8Array(await tracked.arrayBuffer())], [1, 2, 3, 4, 5]);
+  assert.deepEqual(messages.map(message => message.progress), [
+    {asset: 'model', loaded: 2, total: 5},
+    {asset: 'model', loaded: 5, total: 5},
+  ]);
+  assert.ok(messages.every(message => message.id === 7 && message.type === 'progress'));
+});
+
+test('model response progress works when content length is absent', async () => {
+  const {trackDownloadProgress} = await import('../piper-download-progress.mjs');
+  const source = new Response(new Uint8Array([1, 2, 3]));
+  const messages = [];
+  const tracked = trackDownloadProgress(source, {
+    id: 9,
+    asset: 'runtime',
+    file: 'runtime.wasm',
+    postMessage: message => messages.push(message),
+  });
+
+  assert.equal((await tracked.arrayBuffer()).byteLength, 3);
+  assert.deepEqual(messages
+    .map(message => message.progress)
+    .filter(progress => Number.isFinite(progress.loaded)),
+  [{asset: 'runtime', file: 'runtime.wasm', loaded: 3, total: 0}]);
+});
+
+test('runtime asset completion reports that initialization follows the download', async () => {
+  const {trackDownloadProgress} = await import('../piper-download-progress.mjs');
+  const response = new Response(new Uint8Array([1, 2]), {headers: {'content-length': '2'}});
+  const messages = [];
+  const tracked = trackDownloadProgress(response, {
+    id: 10,
+    asset: 'runtime',
+    file: 'ort-wasm-simd-threaded.wasm',
+    postMessage: message => messages.push(message),
+  });
+
+  await tracked.arrayBuffer();
+  assert.deepEqual(messages.at(-1).progress, {
+    stage: 'initializing-runtime',
+    file: 'ort-wasm-simd-threaded.wasm',
+  });
+});

tests/piper-model-routing.test.cjs

@@ -6,11 +6,11 @@ test('Piper model and config requests use the same-origin voice mirror', async (
   const origin = 'https://ai-spel.k3s.k0.nu';
   assert.equal(
     localPiperModelUrl('https://huggingface.co/diffusionstudio/piper-voices/resolve/main/sv/sv_SE/nst/medium/sv_SE-nst-medium.onnx', origin),
-    'https://ai-spel.k3s.k0.nu/tts/piper-wasm-v1/voices/sv/sv_SE/nst/medium/sv_SE-nst-medium.onnx',
+    'https://ai-spel.k3s.k0.nu/static/piper-wasm-v1/voices/sv/sv_SE/nst/medium/sv_SE-nst-medium.onnx',
   );
   assert.equal(
     localPiperModelUrl('https://huggingface.co/diffusionstudio/piper-voices/resolve/main/sv/sv_SE/lisa/medium/sv_SE-lisa-medium.onnx.json', origin),
-    'https://ai-spel.k3s.k0.nu/tts/piper-wasm-v1/voices/sv/sv_SE/lisa/medium/sv_SE-lisa-medium.onnx.json',
+    'https://ai-spel.k3s.k0.nu/static/piper-wasm-v1/voices/sv/sv_SE/lisa/medium/sv_SE-lisa-medium.onnx.json',
   );
 });
 

tests/piper-onnx-config.test.cjs

@@ -0,0 +1,32 @@
+const test = require('node:test');
+const assert = require('node:assert/strict');
+
+test('ONNX Runtime receives an absolute same-origin base URL', async () => {
+  const {createOnnxWasmPaths} = await import('../piper-onnx-config.mjs');
+  const wasmPaths = createOnnxWasmPaths('https://ai-spel.k3s.k0.nu');
+  assert.equal(wasmPaths, 'https://ai-spel.k3s.k0.nu/static/piper-wasm-v1/onnx/');
+  assert.equal(new URL('ort-wasm-simd-threaded.mjs', wasmPaths).href,
+    'https://ai-spel.k3s.k0.nu/static/piper-wasm-v1/onnx/ort-wasm-simd-threaded.mjs');
+});
+
+test('ONNX Runtime base URL rejects an invalid worker origin', async () => {
+  const {createOnnxWasmPaths} = await import('../piper-onnx-config.mjs');
+  assert.throws(() => createOnnxWasmPaths('undefined'), TypeError);
+});
+
+test('Piper and ONNX Runtime stay on one thread even though Piper overwrites ONNX settings', async () => {
+  const {configureOnnxRuntime} = await import('../piper-onnx-config.mjs');
+  const navigatorObject = {hardwareConcurrency: 8};
+  const previousNavigator = globalThis.navigator;
+  Object.defineProperty(globalThis, 'navigator', {configurable: true, value: navigatorObject});
+  const runtime = {env: {wasm: {numThreads: 0, initTimeout: 0}}};
+  try {
+    configureOnnxRuntime(runtime);
+    assert.equal(navigatorObject.hardwareConcurrency, 1);
+    assert.equal(runtime.env.wasm.numThreads, 1);
+    assert.equal(runtime.env.wasm.initTimeout, 60000);
+  } finally {
+    if (previousNavigator === undefined) delete globalThis.navigator;
+    else Object.defineProperty(globalThis, 'navigator', {configurable: true, value: previousNavigator});
+  }
+});

tests/piper-worker-format.test.cjs

@@ -0,0 +1,19 @@
+const test = require('node:test');
+const assert = require('node:assert/strict');
+const {readFile} = require('node:fs/promises');
+const {join} = require('node:path');
+
+test('Piper worker is built and started as an ES module', async () => {
+  const [buildScript, runtime] = await Promise.all([
+    readFile(join(__dirname, '..', 'scripts', 'build-piper.mjs'), 'utf8'),
+    readFile(join(__dirname, '..', 'piper-runtime.js'), 'utf8'),
+  ]);
+  assert.match(buildScript, /outfile: join\(workerRoot, 'piper-worker\.js'\),\s*bundle: true,\s*format: 'esm'/);
+  assert.match(runtime, /new Worker\(workerUrl, \{type: 'module'\}\)/);
+});
+
+test('Piper runtime discards a worker after a failed initialization for a clean retry', async () => {
+  const runtime = await readFile(join(__dirname, '..', 'piper-runtime.js'), 'utf8');
+  assert.match(runtime, /TtsSession keeps a singleton even when ONNX initialization fails/);
+  assert.match(runtime, /worker\?\.terminate\(\);\s*worker = null;/);
+});

tests/voice.test.cjs

@@ -26,6 +26,7 @@ function makeHarness({supported = true, tts = true, piper = false, onTranscript
     constructor(text) { this.text = text; utterances.push(this); }
   }
   const audioSources = [];
+  const piperProgressCallbacks = [];
   class AudioContext {
     constructor() { this.state = 'running'; this.destination = {}; }
     resume() { this.state = 'running'; return Promise.resolve(); }
@@ -49,7 +50,10 @@ function makeHarness({supported = true, tts = true, piper = false, onTranscript
     SpeechSynthesisUtterance: tts ? Utterance : undefined,
     SpeechRecognition: supported ? Recognition : undefined,
     AudioContext,
-    GamePiper: piper ? {synthesize: () => Promise.resolve({arrayBuffer: () => Promise.resolve(new ArrayBuffer(8))})} : undefined,
+    GamePiper: piper ? {synthesize: (_text, onProgress) => {
+      piperProgressCallbacks.push(onProgress);
+      return Promise.resolve({arrayBuffer: () => Promise.resolve(new ArrayBuffer(8))});
+    }} : undefined,
   };
   vm.runInNewContext(source, {window, globalThis: window, Promise, String, Math, Object});
   const controller = window.GameVoice.createVoiceController({
@@ -58,7 +62,7 @@ function makeHarness({supported = true, tts = true, piper = false, onTranscript
     onTranscript,
   });
   return {
-    controller, recognitions, utterances, synthesis, states, timers, audioSources,
+    controller, recognitions, utterances, synthesis, states, timers, audioSources, piperProgressCallbacks,
     runTimer(delay) {
       const match = [...timerDelays.entries()].find(([, value]) => value === delay);
       if (!match) throw new Error(`No timer found with delay ${delay}`);
@@ -250,6 +254,24 @@ test('Piper WASM output uses the unlocked audio context and starts only after lo
   assert.equal(h.controller.getState(), 'IDLE');
 });
 
+test('Piper model download progress is reported as a bounded percentage', async () => {
+  const h = makeHarness({piper: true});
+  h.controller.activateFromUserGesture();
+  h.controller.speak('Ett svar från Piper.');
+  await new Promise(resolve => setImmediate(resolve));
+  h.piperProgressCallbacks[0]({stage: 'loading-model'});
+  assert.match(h.states.at(-1).message, /Ansluter till servern/);
+  h.piperProgressCallbacks[0]({asset: 'runtime', file: 'ort-wasm-simd-threaded.wasm', loaded: 5 * 1024 * 1024, total: 10 * 1024 * 1024});
+  assert.match(h.states.at(-1).message, /ort-wasm-simd-threaded\.wasm: 50 procent/);
+  h.piperProgressCallbacks[0]({loaded: 31_500_000, total: 63_000_000});
+  assert.equal(h.states.at(-1).state, 'PROCESSING');
+  assert.match(h.states.at(-1).message, /50 procent/);
+  h.piperProgressCallbacks[0]({loaded: 90_000_000, total: 63_000_000});
+  assert.match(h.states.at(-1).message, /100 procent/);
+  h.piperProgressCallbacks[0]({loaded: 10 * 1024 * 1024, total: 0});
+  assert.match(h.states.at(-1).message, /10 MB hämtade/);
+});
+
 test('Piper output keeps per-character pitch and rate stable', async () => {
   const h = makeHarness({piper: true});
   h.controller.activateFromUserGesture();

voice.js

@@ -250,13 +250,58 @@
         setState('IDLE', 'Piper tog för lång tid på sig. Dialogen finns kvar som text. Tryck på Läs senaste repliken för att försöka igen.');
       }, 180000);
 
-      let lastProgress = -1;
+      let lastProgressPercent = -1;
+      let lastProgressMegabytes = -1;
+      let lastProgressResource = '';
       const progress = event => {
-        if (token !== generation || !event?.total) return;
+        if (token !== generation) return;
-        const percent = Math.min(100, Math.floor((event.loaded / event.total) * 100));
+        const resource = `${event?.asset || 'model'}:${event?.file || ''}`;
-        if (percent === lastProgress) return;
+        if (resource !== lastProgressResource) {
-        lastProgress = percent;
+          lastProgressResource = resource;
+          lastProgressPercent = -1;
+          lastProgressMegabytes = -1;
+        }
+        if (event?.asset === 'runtime' && Number.isFinite(event.loaded) && event.loaded >= 0) {
+          const file = /^[\w.-]+$/.test(event.file || '') ? event.file : 'WebAssembly-fil';
+          if (Number.isFinite(event.total) && event.total > 0) {
+            const percent = Math.max(0, Math.min(100, Math.floor((event.loaded / event.total) * 100)));
+            if (percent === lastProgressPercent) return;
+            lastProgressPercent = percent;
+            setState('PROCESSING', `Hämtar WebAssembly-filen ${file}: ${percent} procent.`);
+          } else {
+            const megabytes = Math.floor((event.loaded / (1024 * 1024)) * 10) / 10;
+            if (megabytes === lastProgressMegabytes) return;
+            lastProgressMegabytes = megabytes;
+            setState('PROCESSING', `Hämtar WebAssembly-filen ${file}: ${megabytes.toLocaleString('sv-SE')} MB hämtade.`);
+          }
+          return;
+        }
+        const stages = {
+          'loading-runtime': 'Laddar WebAssembly-motorn för svensk uppläsning.',
+          'loading-runtime-assets': `Hämtar WebAssembly-filen ${event?.file || 'från servern'}.`,
+          'reading-runtime-assets': `Läser in WebAssembly-filen ${event?.file || ''}.`,
+          'initializing-runtime': `Startar WebAssembly-motorn efter nedladdning av ${event?.file || 'filen'}.`,
+          'loading-model-info': 'Hämtar information om den svenska röstmodellen.',
+          'loading-model': 'Ansluter till servern för att hämta den svenska röstmodellen.',
+          'reading-model': 'Läser in den svenska röstmodellen.',
+          'model-ready': 'Röstmodellen är inläst. Skapar uppläsningen.',
+        };
+        if (stages[event?.stage]) {
+          setState('PROCESSING', stages[event.stage]);
+          return;
+        }
+        if (!Number.isFinite(event?.loaded) || event.loaded < 0) return;
+        if (Number.isFinite(event.total) && event.total > 0) {
+          const percent = Math.max(0, Math.min(100, Math.floor((event.loaded / event.total) * 100)));
+          if (percent === lastProgressPercent) return;
+          lastProgressPercent = percent;
           setState('PROCESSING', `Hämtar svensk röstmodell till enheten: ${percent} procent.`);
+          return;
+        }
+        const megabytes = Math.floor((event.loaded / (1024 * 1024)) * 10) / 10;
+        if (megabytes === lastProgressMegabytes) return;
+        lastProgressMegabytes = megabytes;
+        setState('PROCESSING', `Hämtar svensk röstmodell till enheten: ${megabytes.toLocaleString('sv-SE')} MB hämtade.`);
       };
 
       Promise.resolve()