@atlaskit/editor-plugin-autocomplete 3.5.0 → 3.6.1

package/dist/cjs/pm-plugins/text-predictor.js

@@ -733,7 +733,8 @@ var unwrapJsonModule = function unwrapJsonModule(mod, shape) {
 };
 var loadVectorsAsync = exports.loadVectorsAsync = /*#__PURE__*/function () {
   var _ref6 = (0, _asyncToGenerator2.default)( /*#__PURE__*/_regenerator.default.mark(function _callee(options) {
-    var url, _wordIndexOuter$index, res, buffer, float32, wordIndexModule, wordIndexOuter, wordIndex, nWords, dim, _t, _t2;
+    var _options$isLocalLLM;
+    var isLocalLLM, url, _wordIndexOuter$index, res, buffer, float32, wordIndexModule, wordIndexOuter, wordIndex, nWords, dim, _t, _t2;
     return _regenerator.default.wrap(function (_context) {
       while (1) switch (_context.prev = _context.next) {
         case 0:
@@ -751,8 +752,11 @@ var loadVectorsAsync = exports.loadVectorsAsync = /*#__PURE__*/function () {
           console.warn('[text-predictor] loadVectorsAsync called without a getBinaryUrl — vectors will not load. Pass getVectorsBinaryUrl via plugin options.');
           return _context.abrupt("return");
         case 2:
+          isLocalLLM = (_options$isLocalLLM = options === null || options === void 0 ? void 0 : options.isLocalLLM) !== null && _options$isLocalLLM !== void 0 ? _options$isLocalLLM : false;
           vectorsLoadStarted = true;
-          (0, _ufo.startExp)(_ufo.EXPERIENCE_NAME.LOAD_VECTORS, 'singleton');
+          (0, _ufo.startExp)(_ufo.EXPERIENCE_NAME.LOAD_VECTORS, 'singleton', {
+            isLocalLLM: isLocalLLM
+          });
           _context.prev = 3;
           _context.next = 4;
           return options.getBinaryUrl();
@@ -765,6 +769,7 @@ var loadVectorsAsync = exports.loadVectorsAsync = /*#__PURE__*/function () {
           _t = _context["catch"](3);
           vectorsLoadStarted = false;
           (0, _ufo.failExp)(_ufo.EXPERIENCE_NAME.LOAD_VECTORS, 'singleton', {
+            isLocalLLM: isLocalLLM,
             errorType: 'resolve_url'
           });
           // eslint-disable-next-line no-console
@@ -782,6 +787,7 @@ var loadVectorsAsync = exports.loadVectorsAsync = /*#__PURE__*/function () {
           }
           vectorsLoadStarted = false;
           (0, _ufo.failExp)(_ufo.EXPERIENCE_NAME.LOAD_VECTORS, 'singleton', {
+            isLocalLLM: isLocalLLM,
             status: res.status,
             errorType: 'http_error'
           });
@@ -816,6 +822,7 @@ var loadVectorsAsync = exports.loadVectorsAsync = /*#__PURE__*/function () {
             dim: dim
           };
           (0, _ufo.succeedExp)(_ufo.EXPERIENCE_NAME.LOAD_VECTORS, 'singleton', {
+            isLocalLLM: isLocalLLM,
             wordCount: nWords,
             dim: dim,
             sizeBytes: float32.byteLength
@@ -835,6 +842,7 @@ var loadVectorsAsync = exports.loadVectorsAsync = /*#__PURE__*/function () {
           _t2 = _context["catch"](6);
           vectorsLoadStarted = false;
           (0, _ufo.failExp)(_ufo.EXPERIENCE_NAME.LOAD_VECTORS, 'singleton', {
+            isLocalLLM: isLocalLLM,
             errorType: 'network'
           });
           // eslint-disable-next-line no-console
@@ -853,19 +861,23 @@ var initVectors = exports.initVectors = function initVectors(store) {
   vectorStore = store;
 };
 var vocabularyLoadPromise;
-var loadDefaultVocabulary = exports.loadDefaultVocabulary = function loadDefaultVocabulary() {
+var loadDefaultVocabulary = exports.loadDefaultVocabulary = function loadDefaultVocabulary(options) {
+  var _options$isLocalLLM2;
   if (isInitialized) {
     return Promise.resolve();
   }
   if (vocabularyLoadPromise) {
     return vocabularyLoadPromise;
   }
+  var isLocalLLM = (_options$isLocalLLM2 = options === null || options === void 0 ? void 0 : options.isLocalLLM) !== null && _options$isLocalLLM2 !== void 0 ? _options$isLocalLLM2 : false;
   vocabularyLoadPromise = (0, _asyncToGenerator2.default)( /*#__PURE__*/_regenerator.default.mark(function _callee2() {
     var _yield$Promise$all, _yield$Promise$all2, vocabularyModule, l3VocabularyModule, vocabularyData, l3VocabularyData, terms, _t3;
     return _regenerator.default.wrap(function (_context2) {
       while (1) switch (_context2.prev = _context2.next) {
         case 0:
-          (0, _ufo.startExp)(_ufo.EXPERIENCE_NAME.LOAD_VOCABULARY, 'singleton');
+          (0, _ufo.startExp)(_ufo.EXPERIENCE_NAME.LOAD_VOCABULARY, 'singleton', {
+            isLocalLLM: isLocalLLM
+          });
           _context2.prev = 1;
           _context2.next = 2;
           return Promise.all([Promise.resolve().then(function () {
@@ -904,6 +916,7 @@ var loadDefaultVocabulary = exports.loadDefaultVocabulary = function loadDefault
             terms: terms
           });
           (0, _ufo.succeedExp)(_ufo.EXPERIENCE_NAME.LOAD_VOCABULARY, 'singleton', {
+            isLocalLLM: isLocalLLM,
             l2WordCount: terms.length,
             l3WordCount: l3VocabularyData.length
           });
@@ -913,6 +926,7 @@ var loadDefaultVocabulary = exports.loadDefaultVocabulary = function loadDefault
           _context2.prev = 4;
           _t3 = _context2["catch"](1);
           (0, _ufo.failExp)(_ufo.EXPERIENCE_NAME.LOAD_VOCABULARY, 'singleton', {
+            isLocalLLM: isLocalLLM,
             errorType: 'parse_error'
           });
           // Allow a later call to retry the load rather than caching the failure.

package/dist/es2019/pm-plugins/autocomplete-plugin.js

@@ -235,12 +235,51 @@ export const createAutocompletePlugin = (options, api) => {
       }
     });
   };
-  const fireSuggestionInsertedAnalytics = ghostText => {
+  const fireLocalModelLoadedAnalytics = info => {
     var _api$analytics2;
+    api === null || api === void 0 ? void 0 : (_api$analytics2 = api.analytics) === null || _api$analytics2 === void 0 ? void 0 : _api$analytics2.actions.fireAnalyticsEvent({
+      action: ACTION.LOCAL_MODEL_LOADED,
+      actionSubject: ACTION_SUBJECT.CONTEXTUAL_TYPEAHEAD,
+      eventType: EVENT_TYPE.TRACK,
+      attributes: {
+        modelId: info.modelId,
+        embeddingModelId: info.embeddingModelId,
+        loadDurationMs: info.loadDurationMs,
+        gpuVendor: info.capabilities.vendor,
+        gpuArchitecture: info.capabilities.architecture
+      }
+    });
+  };
+  const fireLocalModelLoadFailedAnalytics = error => {
+    var _api$analytics3;
+    const {
+      capabilities
+    } = error;
+    api === null || api === void 0 ? void 0 : (_api$analytics3 = api.analytics) === null || _api$analytics3 === void 0 ? void 0 : _api$analytics3.actions.fireAnalyticsEvent({
+      action: ACTION.LOCAL_MODEL_LOAD_FAILED,
+      actionSubject: ACTION_SUBJECT.CONTEXTUAL_TYPEAHEAD,
+      eventType: EVENT_TYPE.TRACK,
+      attributes: {
+        reason: error.reason,
+        message: error.message,
+        modelId: error.modelId,
+        embeddingModelId: error.embeddingModelId,
+        webgpuAvailable: capabilities.available,
+        adapterAvailable: capabilities.adapterAvailable,
+        shaderF16Supported: capabilities.shaderF16Supported,
+        maxBufferSizeMB: capabilities.maxBufferSizeMB,
+        maxStorageBufferBindingSizeMB: capabilities.maxStorageBufferBindingSizeMB,
+        gpuVendor: capabilities.vendor,
+        gpuArchitecture: capabilities.architecture
+      }
+    });
+  };
+  const fireSuggestionInsertedAnalytics = ghostText => {
+    var _api$analytics4;
     const typedLength = lastSuggestionTypedLength;
     const suggestionLength = lastSuggestionLength || typedLength + ghostText.length;
     const kssDelta = suggestionLength - typedLength;
-    api === null || api === void 0 ? void 0 : (_api$analytics2 = api.analytics) === null || _api$analytics2 === void 0 ? void 0 : _api$analytics2.actions.fireAnalyticsEvent({
+    api === null || api === void 0 ? void 0 : (_api$analytics4 = api.analytics) === null || _api$analytics4 === void 0 ? void 0 : _api$analytics4.actions.fireAnalyticsEvent({
       action: ACTION.SUGGESTION_INSERTED,
       actionSubject: ACTION_SUBJECT.CONTEXTUAL_TYPEAHEAD,
       eventType: EVENT_TYPE.TRACK,
@@ -253,7 +292,9 @@ export const createAutocompletePlugin = (options, api) => {
     });
   };
   const slowLaneClient = options !== null && options !== void 0 && options.useLocalModel ? createLocalSlowLaneClient({
-    debounceMs: LOCAL_SLOW_LANE_DEBOUNCE_MS
+    debounceMs: LOCAL_SLOW_LANE_DEBOUNCE_MS,
+    onLoadSuccess: fireLocalModelLoadedAnalytics,
+    onLoadError: fireLocalModelLoadFailedAnalytics
   }) : createSlowLaneClient({
     baseUrl: '',
     debounceMs: NETWORK_SLOW_LANE_DEBOUNCE_MS
@@ -401,9 +442,9 @@ export const createAutocompletePlugin = (options, api) => {
           lastSuggestionLength = typedLength + prediction.length;
           showGhostText(view, prediction, selection.from);
           if (prediction !== lastShownGhostText) {
-            var _api$analytics3;
+            var _api$analytics5;
             lastShownGhostText = prediction;
-            api === null || api === void 0 ? void 0 : (_api$analytics3 = api.analytics) === null || _api$analytics3 === void 0 ? void 0 : _api$analytics3.actions.fireAnalyticsEvent({
+            api === null || api === void 0 ? void 0 : (_api$analytics5 = api.analytics) === null || _api$analytics5 === void 0 ? void 0 : _api$analytics5.actions.fireAnalyticsEvent({
               action: ACTION.SUGGESTION_VIEWED,
               actionSubject: ACTION_SUBJECT.CONTEXTUAL_TYPEAHEAD,
               eventType: EVENT_TYPE.TRACK,
@@ -529,13 +570,17 @@ export const createAutocompletePlugin = (options, api) => {
           return false;
         },
         focus: () => {
-          loadDefaultVocabulary().catch(error => {
+          var _options$useLocalMode, _options$useLocalMode2;
+          loadDefaultVocabulary({
+            isLocalLLM: (_options$useLocalMode = options === null || options === void 0 ? void 0 : options.useLocalModel) !== null && _options$useLocalMode !== void 0 ? _options$useLocalMode : false
+          }).catch(error => {
             logException(error, {
               location: 'editor-plugin-autocomplete/loadDefaultVocabulary'
             });
           });
           loadVectorsAsync({
-            getBinaryUrl: options === null || options === void 0 ? void 0 : options.getVectorsBinaryUrl
+            getBinaryUrl: options === null || options === void 0 ? void 0 : options.getVectorsBinaryUrl,
+            isLocalLLM: (_options$useLocalMode2 = options === null || options === void 0 ? void 0 : options.useLocalModel) !== null && _options$useLocalMode2 !== void 0 ? _options$useLocalMode2 : false
           }).catch(error => {
             logException(error, {
               location: 'editor-plugin-autocomplete/loadVectorsAsync'

package/dist/es2019/pm-plugins/local-slow-lane-client.js

@@ -43,6 +43,19 @@ import { isWordBoundary } from './slow-lane-client';
 
 // Same return type as createSlowLaneClient for drop-in compatibility
 
+/**
+ * Why the local engine failed to load/start.
+ *
+ * The first three are user-machine limitations (WebGPU missing, no compatible
+ * GPU adapter, GPU lacks the `shader-f16` feature the model needs);
+ * `insufficient_memory` is hit when weights don't fit in VRAM. The rest cover
+ * delivery/runtime failures unrelated to hardware.
+ */
+
+/** Snapshot of the machine's WebGPU support, used to explain hardware limits. */
+
+// Minimal WebGPU shape: lib.dom types aren't guaranteed in this build target.
+
 // ─── Constants ───────────────────────────────────────────────────────────────
 
 const DEFAULT_DEBOUNCE_MS = 300;
@@ -470,6 +483,8 @@ export const createLocalSlowLaneClient = (config = {}) => {
     debounceMs = DEFAULT_DEBOUNCE_MS,
     onUpdate,
     onStatus,
+    onLoadError,
+    onLoadSuccess,
     modelId = LOCAL_MLC_CAUSAL_MODEL_ID,
     customModelConfig
   } = config;
@@ -516,16 +531,131 @@ export const createLocalSlowLaneClient = (config = {}) => {
     }
     onStatus === null || onStatus === void 0 ? void 0 : onStatus(message);
   };
+  const bytesToMB = bytes => typeof bytes === 'number' ? Math.round(bytes / (1024 * 1024)) : undefined;
+
+  /**
+   * Inspect the machine's WebGPU support so a load failure can be attributed
+   * to a concrete hardware/browser limitation rather than a generic error.
+   */
+  const probeWebGpuCapabilities = async () => {
+    const gpu = navigator.gpu;
+    if (!gpu) {
+      return {
+        available: false
+      };
+    }
+    try {
+      var _adapter$limits, _adapter$limits2;
+      const adapter = await gpu.requestAdapter();
+      if (!adapter) {
+        return {
+          available: true,
+          adapterAvailable: false
+        };
+      }
+      let vendor;
+      let architecture;
+      try {
+        var _adapter$info, _adapter$requestAdapt;
+        const info = (_adapter$info = adapter.info) !== null && _adapter$info !== void 0 ? _adapter$info : await ((_adapter$requestAdapt = adapter.requestAdapterInfo) === null || _adapter$requestAdapt === void 0 ? void 0 : _adapter$requestAdapt.call(adapter));
+        vendor = (info === null || info === void 0 ? void 0 : info.vendor) || undefined;
+        architecture = (info === null || info === void 0 ? void 0 : info.architecture) || undefined;
+      } catch {
+        // adapter info is best-effort
+      }
+      return {
+        available: true,
+        adapterAvailable: true,
+        shaderF16Supported: adapter.features.has('shader-f16'),
+        maxBufferSizeMB: bytesToMB((_adapter$limits = adapter.limits) === null || _adapter$limits === void 0 ? void 0 : _adapter$limits.maxBufferSize),
+        maxStorageBufferBindingSizeMB: bytesToMB((_adapter$limits2 = adapter.limits) === null || _adapter$limits2 === void 0 ? void 0 : _adapter$limits2.maxStorageBufferBindingSize),
+        vendor,
+        architecture
+      };
+    } catch {
+      return {
+        available: true,
+        adapterAvailable: false
+      };
+    }
+  };
+
+  /** Map an MLC/WebLLM engine-creation error message to a coarse reason. */
+  const classifyEngineError = message => {
+    const lower = message.toLowerCase();
+    if (lower.includes('loading chunk') || lower.includes('dynamically imported module') || lower.includes('dynamic import')) {
+      return 'module_load_failed';
+    }
+    if (lower.includes('out of memory') ||
+    // eslint-disable-next-line require-unicode-regexp
+    /\boom\b/.test(lower) || lower.includes('allocation') || lower.includes('exceeds') || lower.includes('buffer size') || lower.includes('not enough memory')) {
+      return 'insufficient_memory';
+    }
+    // Pre-flight already returns missing_shader_f16 when the feature is absent,
+    // so only match the exact feature token here — not bare 'shader' (compile
+    // errors) or bare 'f16' (present in model ids like q0f16-MLC).
+    if (lower.includes('shader-f16') || lower.includes('shader_f16')) {
+      return 'missing_shader_f16';
+    }
+    if (lower.includes('fetch') || lower.includes('network') || lower.includes('download') || lower.includes('http') || lower.includes('cache')) {
+      return 'model_download_failed';
+    }
+    return 'init_failed';
+  };
+
+  // Canonical, controlled failure descriptions. We never emit the raw engine
+  // error into analytics — it can embed customer-context URLs/paths (HOT-120175)
+  // — so the analytics `message` is always one of these fixed strings.
+  const LOAD_FAILURE_MESSAGE = {
+    webgpu_unavailable: 'WebGPU is not available in this browser',
+    webgpu_no_adapter: 'No compatible WebGPU adapter found',
+    missing_shader_f16: 'GPU does not support the shader-f16 feature',
+    insufficient_memory: 'Insufficient GPU memory to load the model',
+    model_download_failed: 'Failed to download model assets',
+    module_load_failed: 'Failed to load the web-llm runtime module',
+    init_failed: 'Model engine failed to initialise'
+  };
+  const handleLoadFailure = (reason, capabilities, debugDetail) => {
+    ready = false;
+    const message = LOAD_FAILURE_MESSAGE[reason];
+    if (isAutocompleteDebugEnabled()) {
+      // eslint-disable-next-line no-console
+      console.log(`[LocalSlowLane] Engine initialisation failed (${reason}): ${debugDetail !== null && debugDetail !== void 0 ? debugDetail : message}`);
+    }
+    onStatus === null || onStatus === void 0 ? void 0 : onStatus(`Engine initialisation failed: ${message}`);
+    onLoadError === null || onLoadError === void 0 ? void 0 : onLoadError({
+      reason,
+      message,
+      modelId,
+      embeddingModelId: LOCAL_MLC_EMBEDDING_MODEL_ID,
+      capabilities
+    });
+    engineInitPromise = null;
+    initFailed = true;
+  };
   const initEngine = async () => {
+    const capabilities = await probeWebGpuCapabilities();
+
+    // ── Pre-flight: machine limitations short-circuit before the expensive load ──
+    if (!capabilities.available) {
+      handleLoadFailure('webgpu_unavailable', capabilities);
+      return;
+    }
+    if (capabilities.adapterAvailable === false) {
+      handleLoadFailure('webgpu_no_adapter', capabilities);
+      return;
+    }
+    if (capabilities.shaderF16Supported === false) {
+      handleLoadFailure('missing_shader_f16', capabilities);
+      return;
+    }
+    const startTime = performance.now();
     try {
       if (isAutocompleteDebugEnabled()) {
         // eslint-disable-next-line no-console
         console.log(`%c[LocalSlowLane] %c🚀 Initialising MLC engine with models: ${modelId} (LM) + ${LOCAL_MLC_EMBEDDING_MODEL_ID} (embedder)`, 'color: #9c27b0; font-weight: bold;', 'color: inherit;');
       }
       onStatus === null || onStatus === void 0 ? void 0 : onStatus(`Initialising models: ${modelId} + ${LOCAL_MLC_EMBEDDING_MODEL_ID}…`);
-      if (!('gpu' in navigator)) {
-        throw new Error('WebGPU not supported');
-      }
 
       // Fetch the web-llm runtime and the BE-parity lookup tables in parallel;
       // both are dynamically imported so they stay out of the main editor chunk.
@@ -568,6 +698,7 @@ export const createLocalSlowLaneClient = (config = {}) => {
       }
       engine = newEngine;
       ready = true;
+      const loadDurationMs = Math.round(performance.now() - startTime);
       if (isAutocompleteDebugEnabled()) {
         // eslint-disable-next-line no-console
         console.log('%c[LocalSlowLane] %c✅ Both models loaded and ready', 'color: #9c27b0; font-weight: bold;', 'color: #4caf50;');
@@ -579,16 +710,15 @@ export const createLocalSlowLaneClient = (config = {}) => {
         console.log('%c[LocalSlowLane] %c🔢 Embedder     →', 'color: #9c27b0; font-weight: bold;', 'color: #009688; font-weight: bold;', LOCAL_MLC_EMBEDDING_MODEL_ID);
       }
       onStatus === null || onStatus === void 0 ? void 0 : onStatus('Model loaded and ready.');
+      onLoadSuccess === null || onLoadSuccess === void 0 ? void 0 : onLoadSuccess({
+        modelId,
+        embeddingModelId: LOCAL_MLC_EMBEDDING_MODEL_ID,
+        loadDurationMs,
+        capabilities
+      });
     } catch (err) {
       const errorMsg = err instanceof Error ? err.message : String(err);
-      ready = false;
-      if (isAutocompleteDebugEnabled()) {
-        // eslint-disable-next-line no-console
-        console.log(`[LocalSlowLane] Engine initialisation failed: ${errorMsg}`);
-      }
-      onStatus === null || onStatus === void 0 ? void 0 : onStatus(`Engine initialisation failed: ${errorMsg}`);
-      engineInitPromise = null;
-      initFailed = true;
+      handleLoadFailure(classifyEngineError(errorMsg), capabilities, errorMsg);
     }
   };
   const ensureEngineInitialized = () => {

package/dist/es2019/pm-plugins/text-predictor.js

@@ -587,6 +587,7 @@ const unwrapJsonModule = (mod, shape) => {
   return null;
 };
 export const loadVectorsAsync = async options => {
+  var _options$isLocalLLM;
   if (vectorStore || vectorsLoadStarted) {
     return;
   }
@@ -595,14 +596,18 @@ export const loadVectorsAsync = async options => {
     console.warn('[text-predictor] loadVectorsAsync called without a getBinaryUrl — vectors will not load. Pass getVectorsBinaryUrl via plugin options.');
     return;
   }
+  const isLocalLLM = (_options$isLocalLLM = options === null || options === void 0 ? void 0 : options.isLocalLLM) !== null && _options$isLocalLLM !== void 0 ? _options$isLocalLLM : false;
   vectorsLoadStarted = true;
-  startExp(EXPERIENCE_NAME.LOAD_VECTORS, 'singleton');
+  startExp(EXPERIENCE_NAME.LOAD_VECTORS, 'singleton', {
+    isLocalLLM
+  });
   let url;
   try {
     url = await options.getBinaryUrl();
   } catch (e) {
     vectorsLoadStarted = false;
     failExp(EXPERIENCE_NAME.LOAD_VECTORS, 'singleton', {
+      isLocalLLM,
       errorType: 'resolve_url'
     });
     // eslint-disable-next-line no-console
@@ -615,6 +620,7 @@ export const loadVectorsAsync = async options => {
     if (!res.ok) {
       vectorsLoadStarted = false;
       failExp(EXPERIENCE_NAME.LOAD_VECTORS, 'singleton', {
+        isLocalLLM,
         status: res.status,
         errorType: 'http_error'
       });
@@ -643,6 +649,7 @@ export const loadVectorsAsync = async options => {
       dim
     };
     succeedExp(EXPERIENCE_NAME.LOAD_VECTORS, 'singleton', {
+      isLocalLLM,
       wordCount: nWords,
       dim,
       sizeBytes: float32.byteLength
@@ -658,6 +665,7 @@ export const loadVectorsAsync = async options => {
   } catch (e) {
     vectorsLoadStarted = false;
     failExp(EXPERIENCE_NAME.LOAD_VECTORS, 'singleton', {
+      isLocalLLM,
       errorType: 'network'
     });
     // eslint-disable-next-line no-console
@@ -668,15 +676,19 @@ export const initVectors = store => {
   vectorStore = store;
 };
 let vocabularyLoadPromise;
-export const loadDefaultVocabulary = () => {
+export const loadDefaultVocabulary = options => {
+  var _options$isLocalLLM2;
   if (isInitialized) {
     return Promise.resolve();
   }
   if (vocabularyLoadPromise) {
     return vocabularyLoadPromise;
   }
+  const isLocalLLM = (_options$isLocalLLM2 = options === null || options === void 0 ? void 0 : options.isLocalLLM) !== null && _options$isLocalLLM2 !== void 0 ? _options$isLocalLLM2 : false;
   vocabularyLoadPromise = (async () => {
-    startExp(EXPERIENCE_NAME.LOAD_VOCABULARY, 'singleton');
+    startExp(EXPERIENCE_NAME.LOAD_VOCABULARY, 'singleton', {
+      isLocalLLM
+    });
     try {
       // The L2 vocabulary and L3 word list are code-split into their own async
       // chunks so they stay out of the editor's main bundle.
@@ -701,11 +713,13 @@ export const loadDefaultVocabulary = () => {
         terms
       });
       succeedExp(EXPERIENCE_NAME.LOAD_VOCABULARY, 'singleton', {
+        isLocalLLM,
         l2WordCount: terms.length,
         l3WordCount: l3VocabularyData.length
       });
     } catch (e) {
       failExp(EXPERIENCE_NAME.LOAD_VOCABULARY, 'singleton', {
+        isLocalLLM,
         errorType: 'parse_error'
       });
       // Allow a later call to retry the load rather than caching the failure.

package/dist/esm/pm-plugins/autocomplete-plugin.js

@@ -240,12 +240,49 @@ export var createAutocompletePlugin = function createAutocompletePlugin(options,
       }
     });
   };
-  var fireSuggestionInsertedAnalytics = function fireSuggestionInsertedAnalytics(ghostText) {
+  var fireLocalModelLoadedAnalytics = function fireLocalModelLoadedAnalytics(info) {
     var _api$analytics2;
+    api === null || api === void 0 || (_api$analytics2 = api.analytics) === null || _api$analytics2 === void 0 || _api$analytics2.actions.fireAnalyticsEvent({
+      action: ACTION.LOCAL_MODEL_LOADED,
+      actionSubject: ACTION_SUBJECT.CONTEXTUAL_TYPEAHEAD,
+      eventType: EVENT_TYPE.TRACK,
+      attributes: {
+        modelId: info.modelId,
+        embeddingModelId: info.embeddingModelId,
+        loadDurationMs: info.loadDurationMs,
+        gpuVendor: info.capabilities.vendor,
+        gpuArchitecture: info.capabilities.architecture
+      }
+    });
+  };
+  var fireLocalModelLoadFailedAnalytics = function fireLocalModelLoadFailedAnalytics(error) {
+    var _api$analytics3;
+    var capabilities = error.capabilities;
+    api === null || api === void 0 || (_api$analytics3 = api.analytics) === null || _api$analytics3 === void 0 || _api$analytics3.actions.fireAnalyticsEvent({
+      action: ACTION.LOCAL_MODEL_LOAD_FAILED,
+      actionSubject: ACTION_SUBJECT.CONTEXTUAL_TYPEAHEAD,
+      eventType: EVENT_TYPE.TRACK,
+      attributes: {
+        reason: error.reason,
+        message: error.message,
+        modelId: error.modelId,
+        embeddingModelId: error.embeddingModelId,
+        webgpuAvailable: capabilities.available,
+        adapterAvailable: capabilities.adapterAvailable,
+        shaderF16Supported: capabilities.shaderF16Supported,
+        maxBufferSizeMB: capabilities.maxBufferSizeMB,
+        maxStorageBufferBindingSizeMB: capabilities.maxStorageBufferBindingSizeMB,
+        gpuVendor: capabilities.vendor,
+        gpuArchitecture: capabilities.architecture
+      }
+    });
+  };
+  var fireSuggestionInsertedAnalytics = function fireSuggestionInsertedAnalytics(ghostText) {
+    var _api$analytics4;
     var typedLength = lastSuggestionTypedLength;
     var suggestionLength = lastSuggestionLength || typedLength + ghostText.length;
     var kssDelta = suggestionLength - typedLength;
-    api === null || api === void 0 || (_api$analytics2 = api.analytics) === null || _api$analytics2 === void 0 || _api$analytics2.actions.fireAnalyticsEvent({
+    api === null || api === void 0 || (_api$analytics4 = api.analytics) === null || _api$analytics4 === void 0 || _api$analytics4.actions.fireAnalyticsEvent({
       action: ACTION.SUGGESTION_INSERTED,
       actionSubject: ACTION_SUBJECT.CONTEXTUAL_TYPEAHEAD,
       eventType: EVENT_TYPE.TRACK,
@@ -258,7 +295,9 @@ export var createAutocompletePlugin = function createAutocompletePlugin(options,
     });
   };
   var slowLaneClient = options !== null && options !== void 0 && options.useLocalModel ? createLocalSlowLaneClient({
-    debounceMs: LOCAL_SLOW_LANE_DEBOUNCE_MS
+    debounceMs: LOCAL_SLOW_LANE_DEBOUNCE_MS,
+    onLoadSuccess: fireLocalModelLoadedAnalytics,
+    onLoadError: fireLocalModelLoadFailedAnalytics
   }) : createSlowLaneClient({
     baseUrl: '',
     debounceMs: NETWORK_SLOW_LANE_DEBOUNCE_MS
@@ -412,9 +451,9 @@ export var createAutocompletePlugin = function createAutocompletePlugin(options,
           lastSuggestionLength = typedLength + prediction.length;
           showGhostText(view, prediction, selection.from);
           if (prediction !== lastShownGhostText) {
-            var _api$analytics3;
+            var _api$analytics5;
             lastShownGhostText = prediction;
-            api === null || api === void 0 || (_api$analytics3 = api.analytics) === null || _api$analytics3 === void 0 || _api$analytics3.actions.fireAnalyticsEvent({
+            api === null || api === void 0 || (_api$analytics5 = api.analytics) === null || _api$analytics5 === void 0 || _api$analytics5.actions.fireAnalyticsEvent({
               action: ACTION.SUGGESTION_VIEWED,
               actionSubject: ACTION_SUBJECT.CONTEXTUAL_TYPEAHEAD,
               eventType: EVENT_TYPE.TRACK,
@@ -537,13 +576,17 @@ export var createAutocompletePlugin = function createAutocompletePlugin(options,
           return false;
         },
         focus: function focus() {
-          loadDefaultVocabulary().catch(function (error) {
+          var _options$useLocalMode, _options$useLocalMode2;
+          loadDefaultVocabulary({
+            isLocalLLM: (_options$useLocalMode = options === null || options === void 0 ? void 0 : options.useLocalModel) !== null && _options$useLocalMode !== void 0 ? _options$useLocalMode : false
+          }).catch(function (error) {
             logException(error, {
               location: 'editor-plugin-autocomplete/loadDefaultVocabulary'
             });
           });
           loadVectorsAsync({
-            getBinaryUrl: options === null || options === void 0 ? void 0 : options.getVectorsBinaryUrl
+            getBinaryUrl: options === null || options === void 0 ? void 0 : options.getVectorsBinaryUrl,
+            isLocalLLM: (_options$useLocalMode2 = options === null || options === void 0 ? void 0 : options.useLocalModel) !== null && _options$useLocalMode2 !== void 0 ? _options$useLocalMode2 : false
           }).catch(function (error) {
             logException(error, {
               location: 'editor-plugin-autocomplete/loadVectorsAsync'