@atlaskit/editor-plugin-autocomplete 0.2.0 → 0.4.0

package/CHANGELOG.md

@@ -1,5 +1,24 @@
 # @atlaskit/editor-plugin-autocomplete
 
+## 0.4.0
+
+### Minor Changes
+
+- [`603cd44e7b8c3`](https://bitbucket.org/atlassian/atlassian-frontend-monorepo/commits/603cd44e7b8c3) -
+  Fetch vectors through media client
+
+### Patch Changes
+
+- Updated dependencies
+
+## 0.3.0
+
+### Minor Changes
+
+- [`87965237565b6`](https://bitbucket.org/atlassian/atlassian-frontend-monorepo/commits/87965237565b6) -
+  [ux] Add autocomplete plugin to inline comment editor behind
+  `platform_editor_ai_autocomplete_conf_comments` experiment
+
 ## 0.2.0
 
 ### Minor Changes

package/afm-cc/tsconfig.json

@@ -6,10 +6,7 @@
 		"rootDir": "../",
 		"composite": true,
 		"noCheck": true,
-		// resolveJsonModule must remain false: this package contains LFS-tracked JSON data files
-		// that appear as pointer text (not valid JSON) in CI environments where LFS is skipped.
-		// Enabling resolveJsonModule would cause TS to attempt parsing these files and fail.
-		"resolveJsonModule": false
+		"resolveJsonModule": true
 	},
 	"include": [
 		"../src/**/*.ts",

package/afm-jira/tsconfig.json

@@ -6,10 +6,7 @@
 		"rootDir": "../",
 		"composite": true,
 		"noCheck": true,
-		// resolveJsonModule must remain false: this package contains LFS-tracked JSON data files
-		// that appear as pointer text (not valid JSON) in CI environments where LFS is skipped.
-		// Enabling resolveJsonModule would cause TS to attempt parsing these files and fail.
-		"resolveJsonModule": false
+		"resolveJsonModule": true
 	},
 	"include": [
 		"../src/**/*.ts",

package/afm-products/tsconfig.json

@@ -6,10 +6,7 @@
 		"rootDir": "../",
 		"composite": true,
 		"noCheck": true,
-		// resolveJsonModule must remain false: this package contains LFS-tracked JSON data files
-		// that appear as pointer text (not valid JSON) in CI environments where LFS is skipped.
-		// Enabling resolveJsonModule would cause TS to attempt parsing these files and fail.
-		"resolveJsonModule": false
+		"resolveJsonModule": true
 	},
 	"include": [
 		"../src/**/*.ts",

package/build/tsconfig.json

@@ -2,12 +2,7 @@
 	"extends": "../tsconfig",
 	"compilerOptions": {
 		"paths": {},
-		// resolveJsonModule must remain false: this package contains LFS-tracked JSON data files
-		// that appear as pointer text (not valid JSON) in CI environments where LFS is skipped.
-		// We also intentionally do not set "target": "es5" here (unlike the standard generated
-		// build tsconfig) because the ESLint require-unicode-regexp rule enforces the 'u' flag
-		// on regexes, which requires es6+ target to compile.
-		"resolveJsonModule": false
+		"resolveJsonModule": true
 	},
 	"include": ["../src/**/*.ts", "../src/**/*.tsx"],
 	"exclude": [

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

@@ -214,14 +214,16 @@ var createAutocompletePlugin = exports.createAutocompletePlugin = function creat
       return;
     }
     var lastChar = newText[newText.length - 1];
-    if (!/[\t-\r !,\.:;\?\xA0\u1680\u2000-\u200A\u2028\u2029\u202F\u205F\u3000\uFEFF]/.test(lastChar)) {
+    // eslint-disable-next-line require-unicode-regexp
+    if (!/[\s.,;:!?]/.test(lastChar)) {
       return;
     }
 
     // Only fire if the previous state did not already end on a boundary,
     // so we don't double-count when multiple boundary chars are inserted.
     var prevLastChar = prevText[prevText.length - 1];
-    if (prevLastChar && /[\t-\r !,\.:;\?\xA0\u1680\u2000-\u200A\u2028\u2029\u202F\u205F\u3000\uFEFF]/.test(prevLastChar)) {
+    // eslint-disable-next-line require-unicode-regexp
+    if (prevLastChar && /[\s.,;:!?]/.test(prevLastChar)) {
       return;
     }
     var beforeBoundary = newText.slice(0, -1).trimEnd();
@@ -299,7 +301,9 @@ var createAutocompletePlugin = exports.createAutocompletePlugin = function creat
         },
         focus: function focus() {
           (0, _textPredictor.loadDefaultVocabulary)();
-          (0, _textPredictor.loadVectorsAsync)().catch(function () {});
+          (0, _textPredictor.loadVectorsAsync)({
+            getBinaryUrl: options === null || options === void 0 ? void 0 : options.getVectorsBinaryUrl
+          }).catch(function () {});
           if (!hasIngestedPage) {
             hasIngestedPage = true;
             if (options !== null && options !== void 0 && options.getContext) {

package/dist/cjs/pm-plugins/scoring-pipeline.js

@@ -16,9 +16,7 @@ function _arrayLikeToArray(r, a) { (null == a || a > r.length) && (a = r.length)
  *
  * Operates synchronously on pre-loaded data. Each stage gracefully degrades
  * when its required data isn't available (cold → warm → full warm).
- */ // resolveJsonModule is disabled for this package (see tsconfig.json) to prevent
-// TypeScript from parsing LFS pointer files during CI typecheck. JSON imports
-// are typed via the '*.json' declaration in typings.d.ts.
+ */
 // ─── Types ──────────────────────────────────────────────────
 
 /** Metadata returned by the grammar filter for debug logging in the caller. */

package/dist/cjs/pm-plugins/slow-lane-client.js

@@ -25,7 +25,8 @@ var _asyncToGenerator2 = _interopRequireDefault(require("@babel/runtime/helpers/
 
 // ─── Constants ───────────────────────────────────────────────────────────────
 
-var WORD_BOUNDARY_CHARS = /[\t-\r !,\.:;\?\xA0\u1680\u2000-\u200A\u2028\u2029\u202F\u205F\u3000\uFEFF]/;
+// eslint-disable-next-line require-unicode-regexp
+var WORD_BOUNDARY_CHARS = /[\s.,;:!?]/;
 var DEFAULT_DEBOUNCE_MS = 300;
 
 /**
@@ -75,6 +76,7 @@ var createSlowLaneClient = exports.createSlowLaneClient = function createSlowLan
             }
             return _context.abrupt("return");
           case 2:
+            // eslint-disable-next-line require-unicode-regexp
             url = "".concat(baseUrl.replace(/\/$/, '')).concat(endpoint);
             payload = {
               text: text,

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

@@ -37,7 +37,8 @@ function _arrayLikeToArray(r, a) { (null == a || a > r.length) && (a = r.length)
 // import { rankCandidates, isGrammarAllowed } from './scoring-pipeline';
 // ─── Constants ───────────────────────────────────────────────────────────────
 
-var PUNCTUATION_BOUNDARY_REGEX = /^[!"'-\),\.:;\?\[\]`\{\}]+|[!"'-\),\.:;\?\[\]`\{\}]+$/g;
+// eslint-disable-next-line require-unicode-regexp
+var PUNCTUATION_BOUNDARY_REGEX = /^[.,;:!?()\[\]{}"'`]+|[.,;:!?()\[\]{}"'`]+$/g;
 var MIN_PREFIX_LENGTH = 3;
 var MAX_CANDIDATES = 200;
 var CONTEXT_WORDS = 10;
@@ -311,8 +312,8 @@ var getContextVectorForScoring = function getContextVectorForScoring(textBefore)
 };
 var tokenize = function tokenize(text) {
   var tokens = [];
-  // eslint-disable-next-line @atlassian/perf-linting/no-expensive-split-replace
-  var _iterator7 = _createForOfIteratorHelper(text.toLowerCase().split(/[\t-\r \xA0\u1680\u2000-\u200A\u2028\u2029\u202F\u205F\u3000\uFEFF]+/)),
+  // eslint-disable-next-line require-unicode-regexp, @atlassian/perf-linting/no-expensive-split-replace
+  var _iterator7 = _createForOfIteratorHelper(text.toLowerCase().split(/\s+/)),
     _step7;
   try {
     for (_iterator7.s(); !(_step7 = _iterator7.n()).done;) {
@@ -332,13 +333,15 @@ var tokenize = function tokenize(text) {
 };
 var extractPreviousWord = function extractPreviousWord(text) {
   // 1. Split the text by newlines or punctuation (. ? !)
-  var sentences = text.split(/[\n!\.\?]+/);
+  // eslint-disable-next-line require-unicode-regexp
+  var sentences = text.split(/[\n.?!]+/);
 
   // 2. Only look at the current sentence/line the user is typing in
   var currentSentence = sentences[sentences.length - 1];
 
   // 3. Extract the previous word as normal
-  var words = currentSentence.trimEnd().split(/[\t-\r \xA0\u1680\u2000-\u200A\u2028\u2029\u202F\u205F\u3000\uFEFF]+/);
+  // eslint-disable-next-line require-unicode-regexp
+  var words = currentSentence.trimEnd().split(/\s+/);
   return words.length >= 2 ? words[words.length - 2] : '';
 };
 
@@ -489,7 +492,8 @@ var predict = exports.predict = function predict(textBefore) {
   // }
 
   // ── Step 2: Prefix completion (≥3 chars typed) ──────────────────────────
-  if (textBefore.length > 0 && /[\t-\r \xA0\u1680\u2000-\u200A\u2028\u2029\u202F\u205F\u3000\uFEFF]$/.test(textBefore)) {
+  // eslint-disable-next-line require-unicode-regexp
+  if (textBefore.length > 0 && /\s$/.test(textBefore)) {
     return null;
   }
   var trimmed = textBefore.trimEnd();
@@ -689,10 +693,8 @@ var predict = exports.predict = function predict(textBefore) {
 
 // ─── Data Loading ────────────────────────────────────────────────────────────
 
-var DEFAULT_VECTORS_URL = '/data/word-vectors_10k.bin';
 var loadVectorsAsync = exports.loadVectorsAsync = /*#__PURE__*/function () {
   var _ref4 = (0, _asyncToGenerator2.default)( /*#__PURE__*/_regenerator.default.mark(function _callee(options) {
-    var _options$vectorsUrl;
     var url, res, buffer, float32, wordIndex, nWords, dim;
     return _regenerator.default.wrap(function _callee$(_context) {
       while (1) switch (_context.prev = _context.next) {
@@ -703,25 +705,47 @@ var loadVectorsAsync = exports.loadVectorsAsync = /*#__PURE__*/function () {
           }
           return _context.abrupt("return");
         case 2:
+          if (options !== null && options !== void 0 && options.getBinaryUrl) {
+            _context.next = 5;
+            break;
+          }
+          // eslint-disable-next-line no-console
+          console.warn('[text-predictor] loadVectorsAsync called without a getBinaryUrl — vectors will not load. Pass getVectorsBinaryUrl via plugin options.');
+          return _context.abrupt("return");
+        case 5:
           vectorsLoadStarted = true;
-          url = (_options$vectorsUrl = options === null || options === void 0 ? void 0 : options.vectorsUrl) !== null && _options$vectorsUrl !== void 0 ? _options$vectorsUrl : DEFAULT_VECTORS_URL;
-          _context.prev = 4;
-          _context.next = 7;
+          _context.prev = 6;
+          _context.next = 9;
+          return options.getBinaryUrl();
+        case 9:
+          url = _context.sent;
+          _context.next = 17;
+          break;
+        case 12:
+          _context.prev = 12;
+          _context.t0 = _context["catch"](6);
+          vectorsLoadStarted = false;
+          // eslint-disable-next-line no-console
+          console.warn('[text-predictor] Failed to resolve vectors URL:', _context.t0);
+          return _context.abrupt("return");
+        case 17:
+          _context.prev = 17;
+          _context.next = 20;
           return fetch(url);
-        case 7:
+        case 20:
           res = _context.sent;
           if (res.ok) {
-            _context.next = 12;
+            _context.next = 25;
             break;
           }
           vectorsLoadStarted = false;
           // eslint-disable-next-line no-console
           console.warn("[text-predictor] Failed to load vectors: ".concat(res.status));
           return _context.abrupt("return");
-        case 12:
-          _context.next = 14;
+        case 25:
+          _context.next = 27;
           return res.arrayBuffer();
-        case 14:
+        case 27:
           buffer = _context.sent;
           float32 = new Float32Array(buffer);
           wordIndex = _word_index_10k.default;
@@ -741,19 +765,19 @@ var loadVectorsAsync = exports.loadVectorsAsync = /*#__PURE__*/function () {
               sizeBytes: float32.byteLength
             });
           }
-          _context.next = 28;
+          _context.next = 41;
           break;
-        case 24:
-          _context.prev = 24;
-          _context.t0 = _context["catch"](4);
+        case 37:
+          _context.prev = 37;
+          _context.t1 = _context["catch"](17);
           vectorsLoadStarted = false;
           // eslint-disable-next-line no-console
-          console.warn('[text-predictor] Failed to load vectors:', _context.t0);
-        case 28:
+          console.warn('[text-predictor] Failed to load vectors:', _context.t1);
+        case 41:
         case "end":
           return _context.stop();
       }
-    }, _callee, null, [[4, 24]]);
+    }, _callee, null, [[6, 12], [17, 37]]);
   }));
   return function loadVectorsAsync(_x) {
     return _ref4.apply(this, arguments);

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

@@ -212,14 +212,16 @@ export const createAutocompletePlugin = options => {
       return;
     }
     const lastChar = newText[newText.length - 1];
-    if (!/[\s.,;:!?]/u.test(lastChar)) {
+    // eslint-disable-next-line require-unicode-regexp
+    if (!/[\s.,;:!?]/.test(lastChar)) {
       return;
     }
 
     // Only fire if the previous state did not already end on a boundary,
     // so we don't double-count when multiple boundary chars are inserted.
     const prevLastChar = prevText[prevText.length - 1];
-    if (prevLastChar && /[\s.,;:!?]/u.test(prevLastChar)) {
+    // eslint-disable-next-line require-unicode-regexp
+    if (prevLastChar && /[\s.,;:!?]/.test(prevLastChar)) {
       return;
     }
     const beforeBoundary = newText.slice(0, -1).trimEnd();
@@ -300,7 +302,9 @@ export const createAutocompletePlugin = options => {
         },
         focus: () => {
           loadDefaultVocabulary();
-          loadVectorsAsync().catch(() => {});
+          loadVectorsAsync({
+            getBinaryUrl: options === null || options === void 0 ? void 0 : options.getVectorsBinaryUrl
+          }).catch(() => {});
           if (!hasIngestedPage) {
             hasIngestedPage = true;
             if (options !== null && options !== void 0 && options.getContext) {

package/dist/es2019/pm-plugins/scoring-pipeline.js

@@ -5,9 +5,6 @@
  * when its required data isn't available (cold → warm → full warm).
  */
 
-// resolveJsonModule is disabled for this package (see tsconfig.json) to prevent
-// TypeScript from parsing LFS pointer files during CI typecheck. JSON imports
-// are typed via the '*.json' declaration in typings.d.ts.
 import posTagsData from './data/combined_l2_l3_pos_tags.json';
 import ghostPosTagsData from './data/ghost_pos_tags.json';
 import grammarTransitionsData from './data/grammar_transitions_10k.json';

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

@@ -15,7 +15,8 @@
 
 // ─── Constants ───────────────────────────────────────────────────────────────
 
-const WORD_BOUNDARY_CHARS = /[\s.,;:!?]/u;
+// eslint-disable-next-line require-unicode-regexp
+const WORD_BOUNDARY_CHARS = /[\s.,;:!?]/;
 const DEFAULT_DEBOUNCE_MS = 300;
 
 /**
@@ -55,7 +56,9 @@ export const createSlowLaneClient = config => {
     if (!text || text.trim().length === 0) {
       return;
     }
-    const url = `${baseUrl.replace(/\/$/u, '')}${endpoint}`;
+
+    // eslint-disable-next-line require-unicode-regexp
+    const url = `${baseUrl.replace(/\/$/, '')}${endpoint}`;
     const payload = {
       text,
       session_id: sessionId

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

@@ -26,7 +26,8 @@ import { getStoredContextVector, getStoredLmLogits } from './slow-lane-client';
 
 // ─── Constants ───────────────────────────────────────────────────────────────
 
-const PUNCTUATION_BOUNDARY_REGEX = /^[.,;:!?()\[\]{}"'`]+|[.,;:!?()\[\]{}"'`]+$/gu;
+// eslint-disable-next-line require-unicode-regexp
+const PUNCTUATION_BOUNDARY_REGEX = /^[.,;:!?()\[\]{}"'`]+|[.,;:!?()\[\]{}"'`]+$/g;
 const MIN_PREFIX_LENGTH = 3;
 const MAX_CANDIDATES = 200;
 const CONTEXT_WORDS = 10;
@@ -236,8 +237,8 @@ const getContextVectorForScoring = textBefore => {
 };
 const tokenize = text => {
   const tokens = [];
-  // eslint-disable-next-line @atlassian/perf-linting/no-expensive-split-replace
-  for (const raw of text.toLowerCase().split(/\s+/u)) {
+  // eslint-disable-next-line require-unicode-regexp, @atlassian/perf-linting/no-expensive-split-replace
+  for (const raw of text.toLowerCase().split(/\s+/)) {
     // eslint-disable-next-line @atlassian/perf-linting/no-expensive-split-replace
     const clean = raw.replace(PUNCTUATION_BOUNDARY_REGEX, '');
     if (clean.length >= 2) {
@@ -248,13 +249,15 @@ const tokenize = text => {
 };
 const extractPreviousWord = text => {
   // 1. Split the text by newlines or punctuation (. ? !)
-  const sentences = text.split(/[\n.?!]+/u);
+  // eslint-disable-next-line require-unicode-regexp
+  const sentences = text.split(/[\n.?!]+/);
 
   // 2. Only look at the current sentence/line the user is typing in
   const currentSentence = sentences[sentences.length - 1];
 
   // 3. Extract the previous word as normal
-  const words = currentSentence.trimEnd().split(/\s+/u);
+  // eslint-disable-next-line require-unicode-regexp
+  const words = currentSentence.trimEnd().split(/\s+/);
   return words.length >= 2 ? words[words.length - 2] : '';
 };
 
@@ -387,7 +390,8 @@ export const predict = textBefore => {
   // }
 
   // ── Step 2: Prefix completion (≥3 chars typed) ──────────────────────────
-  if (textBefore.length > 0 && /\s$/u.test(textBefore)) {
+  // eslint-disable-next-line require-unicode-regexp
+  if (textBefore.length > 0 && /\s$/.test(textBefore)) {
     return null;
   }
   const trimmed = textBefore.trimEnd();
@@ -561,14 +565,25 @@ export const predict = textBefore => {
 
 // ─── Data Loading ────────────────────────────────────────────────────────────
 
-const DEFAULT_VECTORS_URL = '/data/word-vectors_10k.bin';
 export const loadVectorsAsync = async options => {
-  var _options$vectorsUrl;
   if (vectorStore || vectorsLoadStarted) {
     return;
   }
+  if (!(options !== null && options !== void 0 && options.getBinaryUrl)) {
+    // eslint-disable-next-line no-console
+    console.warn('[text-predictor] loadVectorsAsync called without a getBinaryUrl — vectors will not load. Pass getVectorsBinaryUrl via plugin options.');
+    return;
+  }
   vectorsLoadStarted = true;
-  const url = (_options$vectorsUrl = options === null || options === void 0 ? void 0 : options.vectorsUrl) !== null && _options$vectorsUrl !== void 0 ? _options$vectorsUrl : DEFAULT_VECTORS_URL;
+  let url;
+  try {
+    url = await options.getBinaryUrl();
+  } catch (e) {
+    vectorsLoadStarted = false;
+    // eslint-disable-next-line no-console
+    console.warn('[text-predictor] Failed to resolve vectors URL:', e);
+    return;
+  }
   try {
     const res = await fetch(url);
     if (!res.ok) {

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

@@ -207,14 +207,16 @@ export var createAutocompletePlugin = function createAutocompletePlugin(options)
       return;
     }
     var lastChar = newText[newText.length - 1];
-    if (!/[\t-\r !,\.:;\?\xA0\u1680\u2000-\u200A\u2028\u2029\u202F\u205F\u3000\uFEFF]/.test(lastChar)) {
+    // eslint-disable-next-line require-unicode-regexp
+    if (!/[\s.,;:!?]/.test(lastChar)) {
       return;
     }
 
     // Only fire if the previous state did not already end on a boundary,
     // so we don't double-count when multiple boundary chars are inserted.
     var prevLastChar = prevText[prevText.length - 1];
-    if (prevLastChar && /[\t-\r !,\.:;\?\xA0\u1680\u2000-\u200A\u2028\u2029\u202F\u205F\u3000\uFEFF]/.test(prevLastChar)) {
+    // eslint-disable-next-line require-unicode-regexp
+    if (prevLastChar && /[\s.,;:!?]/.test(prevLastChar)) {
       return;
     }
     var beforeBoundary = newText.slice(0, -1).trimEnd();
@@ -292,7 +294,9 @@ export var createAutocompletePlugin = function createAutocompletePlugin(options)
         },
         focus: function focus() {
           loadDefaultVocabulary();
-          loadVectorsAsync().catch(function () {});
+          loadVectorsAsync({
+            getBinaryUrl: options === null || options === void 0 ? void 0 : options.getVectorsBinaryUrl
+          }).catch(function () {});
           if (!hasIngestedPage) {
             hasIngestedPage = true;
             if (options !== null && options !== void 0 && options.getContext) {

package/dist/esm/pm-plugins/scoring-pipeline.js

@@ -9,9 +9,6 @@ function _arrayLikeToArray(r, a) { (null == a || a > r.length) && (a = r.length)
  * when its required data isn't available (cold → warm → full warm).
  */
 
-// resolveJsonModule is disabled for this package (see tsconfig.json) to prevent
-// TypeScript from parsing LFS pointer files during CI typecheck. JSON imports
-// are typed via the '*.json' declaration in typings.d.ts.
 import posTagsData from './data/combined_l2_l3_pos_tags.json';
 import ghostPosTagsData from './data/ghost_pos_tags.json';
 import grammarTransitionsData from './data/grammar_transitions_10k.json';

package/dist/esm/pm-plugins/slow-lane-client.js

@@ -18,7 +18,8 @@ import _regeneratorRuntime from "@babel/runtime/regenerator";
 
 // ─── Constants ───────────────────────────────────────────────────────────────
 
-var WORD_BOUNDARY_CHARS = /[\t-\r !,\.:;\?\xA0\u1680\u2000-\u200A\u2028\u2029\u202F\u205F\u3000\uFEFF]/;
+// eslint-disable-next-line require-unicode-regexp
+var WORD_BOUNDARY_CHARS = /[\s.,;:!?]/;
 var DEFAULT_DEBOUNCE_MS = 300;
 
 /**
@@ -68,6 +69,7 @@ export var createSlowLaneClient = function createSlowLaneClient(config) {
             }
             return _context.abrupt("return");
           case 2:
+            // eslint-disable-next-line require-unicode-regexp
             url = "".concat(baseUrl.replace(/\/$/, '')).concat(endpoint);
             payload = {
               text: text,

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

@@ -34,7 +34,8 @@ import { getStoredContextVector, getStoredLmLogits } from './slow-lane-client';
 
 // ─── Constants ───────────────────────────────────────────────────────────────
 
-var PUNCTUATION_BOUNDARY_REGEX = /^[!"'-\),\.:;\?\[\]`\{\}]+|[!"'-\),\.:;\?\[\]`\{\}]+$/g;
+// eslint-disable-next-line require-unicode-regexp
+var PUNCTUATION_BOUNDARY_REGEX = /^[.,;:!?()\[\]{}"'`]+|[.,;:!?()\[\]{}"'`]+$/g;
 var MIN_PREFIX_LENGTH = 3;
 var MAX_CANDIDATES = 200;
 var CONTEXT_WORDS = 10;
@@ -308,8 +309,8 @@ var getContextVectorForScoring = function getContextVectorForScoring(textBefore)
 };
 var tokenize = function tokenize(text) {
   var tokens = [];
-  // eslint-disable-next-line @atlassian/perf-linting/no-expensive-split-replace
-  var _iterator7 = _createForOfIteratorHelper(text.toLowerCase().split(/[\t-\r \xA0\u1680\u2000-\u200A\u2028\u2029\u202F\u205F\u3000\uFEFF]+/)),
+  // eslint-disable-next-line require-unicode-regexp, @atlassian/perf-linting/no-expensive-split-replace
+  var _iterator7 = _createForOfIteratorHelper(text.toLowerCase().split(/\s+/)),
     _step7;
   try {
     for (_iterator7.s(); !(_step7 = _iterator7.n()).done;) {
@@ -329,13 +330,15 @@ var tokenize = function tokenize(text) {
 };
 var extractPreviousWord = function extractPreviousWord(text) {
   // 1. Split the text by newlines or punctuation (. ? !)
-  var sentences = text.split(/[\n!\.\?]+/);
+  // eslint-disable-next-line require-unicode-regexp
+  var sentences = text.split(/[\n.?!]+/);
 
   // 2. Only look at the current sentence/line the user is typing in
   var currentSentence = sentences[sentences.length - 1];
 
   // 3. Extract the previous word as normal
-  var words = currentSentence.trimEnd().split(/[\t-\r \xA0\u1680\u2000-\u200A\u2028\u2029\u202F\u205F\u3000\uFEFF]+/);
+  // eslint-disable-next-line require-unicode-regexp
+  var words = currentSentence.trimEnd().split(/\s+/);
   return words.length >= 2 ? words[words.length - 2] : '';
 };
 
@@ -486,7 +489,8 @@ export var predict = function predict(textBefore) {
   // }
 
   // ── Step 2: Prefix completion (≥3 chars typed) ──────────────────────────
-  if (textBefore.length > 0 && /[\t-\r \xA0\u1680\u2000-\u200A\u2028\u2029\u202F\u205F\u3000\uFEFF]$/.test(textBefore)) {
+  // eslint-disable-next-line require-unicode-regexp
+  if (textBefore.length > 0 && /\s$/.test(textBefore)) {
     return null;
   }
   var trimmed = textBefore.trimEnd();
@@ -686,10 +690,8 @@ export var predict = function predict(textBefore) {
 
 // ─── Data Loading ────────────────────────────────────────────────────────────
 
-var DEFAULT_VECTORS_URL = '/data/word-vectors_10k.bin';
 export var loadVectorsAsync = /*#__PURE__*/function () {
   var _ref4 = _asyncToGenerator( /*#__PURE__*/_regeneratorRuntime.mark(function _callee(options) {
-    var _options$vectorsUrl;
     var url, res, buffer, float32, wordIndex, nWords, dim;
     return _regeneratorRuntime.wrap(function _callee$(_context) {
       while (1) switch (_context.prev = _context.next) {
@@ -700,25 +702,47 @@ export var loadVectorsAsync = /*#__PURE__*/function () {
           }
           return _context.abrupt("return");
         case 2:
+          if (options !== null && options !== void 0 && options.getBinaryUrl) {
+            _context.next = 5;
+            break;
+          }
+          // eslint-disable-next-line no-console
+          console.warn('[text-predictor] loadVectorsAsync called without a getBinaryUrl — vectors will not load. Pass getVectorsBinaryUrl via plugin options.');
+          return _context.abrupt("return");
+        case 5:
           vectorsLoadStarted = true;
-          url = (_options$vectorsUrl = options === null || options === void 0 ? void 0 : options.vectorsUrl) !== null && _options$vectorsUrl !== void 0 ? _options$vectorsUrl : DEFAULT_VECTORS_URL;
-          _context.prev = 4;
-          _context.next = 7;
+          _context.prev = 6;
+          _context.next = 9;
+          return options.getBinaryUrl();
+        case 9:
+          url = _context.sent;
+          _context.next = 17;
+          break;
+        case 12:
+          _context.prev = 12;
+          _context.t0 = _context["catch"](6);
+          vectorsLoadStarted = false;
+          // eslint-disable-next-line no-console
+          console.warn('[text-predictor] Failed to resolve vectors URL:', _context.t0);
+          return _context.abrupt("return");
+        case 17:
+          _context.prev = 17;
+          _context.next = 20;
           return fetch(url);
-        case 7:
+        case 20:
           res = _context.sent;
           if (res.ok) {
-            _context.next = 12;
+            _context.next = 25;
             break;
           }
           vectorsLoadStarted = false;
           // eslint-disable-next-line no-console
           console.warn("[text-predictor] Failed to load vectors: ".concat(res.status));
           return _context.abrupt("return");
-        case 12:
-          _context.next = 14;
+        case 25:
+          _context.next = 27;
           return res.arrayBuffer();
-        case 14:
+        case 27:
           buffer = _context.sent;
           float32 = new Float32Array(buffer);
           wordIndex = wordIndexData;
@@ -738,19 +762,19 @@ export var loadVectorsAsync = /*#__PURE__*/function () {
               sizeBytes: float32.byteLength
             });
           }
-          _context.next = 28;
+          _context.next = 41;
           break;
-        case 24:
-          _context.prev = 24;
-          _context.t0 = _context["catch"](4);
+        case 37:
+          _context.prev = 37;
+          _context.t1 = _context["catch"](17);
           vectorsLoadStarted = false;
           // eslint-disable-next-line no-console
-          console.warn('[text-predictor] Failed to load vectors:', _context.t0);
-        case 28:
+          console.warn('[text-predictor] Failed to load vectors:', _context.t1);
+        case 41:
         case "end":
           return _context.stop();
       }
-    }, _callee, null, [[4, 24]]);
+    }, _callee, null, [[6, 12], [17, 37]]);
   }));
   return function loadVectorsAsync(_x) {
     return _ref4.apply(this, arguments);

package/dist/types/pm-plugins/autocomplete-plugin.d.ts

@@ -32,5 +32,11 @@ export interface AutocompletePluginOptions {
      * word-frequency boosting. Called lazily so the preset can remain synchronous.
      */
     getContext?: () => Promise<AutocompleteContext | undefined>;
+    /**
+     * Async function that resolves to a URL for the word vectors binary file.
+     * When provided, this takes precedence over the bundled asset URL.
+     * Use this to serve vectors from a CDN or media service in production.
+     */
+    getVectorsBinaryUrl?: () => Promise<string>;
 }
 export declare const createAutocompletePlugin: (options?: AutocompletePluginOptions) => SafePlugin<AutocompletePluginState>;

package/dist/types/pm-plugins/text-predictor.d.ts

@@ -83,7 +83,7 @@ export declare const incrementSessionFreq: (word: string) => void;
 export declare const ingestDocumentPage: (pageContent: string | undefined) => void;
 export declare const predict: (textBefore: string) => string | null;
 export declare const loadVectorsAsync: (options?: {
-    vectorsUrl?: string;
+    getBinaryUrl?: () => Promise<string>;
 }) => Promise<void>;
 export declare const initVectors: (store: VectorStore) => void;
 export declare const loadDefaultVocabulary: () => void;