semantic titles ai assist

app/capture.js

@@ -3,7 +3,6 @@
 const path = require('node:path');
 const { spawn, execFileSync } = require('node:child_process');
 const { desktopCapturer, screen, BrowserWindow, nativeImage, Tray, Menu } = require('electron');
-const { expandPlaceholders } = require('../core/placeholders');
 const raster = require('../core/raster');
 const { encodePng } = require('../core/png');
 const {
@@ -14,6 +13,7 @@ const {
   DEFAULT_START_SLACK_MS,
 } = require('./click-frames');
 const { physicalToDip } = require('./coords');
+const { DEFAULT_CAPTURE_TITLES } = require('../core/text-intel');
 
 /**
  * Capture service: full-screen, active-window, and region capture, plus a
@@ -114,6 +114,7 @@ class CaptureService {
     getWindow,
     notify,
     screenApi = screen,
+    textIntel = null,
   }) {
     this.store = store;
     this.settings = settings;
@@ -122,6 +123,7 @@ class CaptureService {
     // Injectable for tests; the click/coordinate paths must never reach for
     // the global `screen` directly so coordinate handling stays testable.
     this.screen = screenApi;
+    this.textIntel = textIntel;
     this.session = null; // { guideId, paused, count, intervalSec }
     this.intervalTimer = null;
     this.clickWatcher = null;
@@ -133,6 +135,7 @@ class CaptureService {
     this.clickWatcherErrTail = '';
     this.linuxEvent = null; // event block currently being parsed
     this.pendingRawClick = null; // raw press waiting for its coordinate twin
+    this.pendingClickOsPoint = null;
     this.clickQueue = Promise.resolve();
     this.frameLoopInFlight = false;
     this.frameLoopGrabStartedAt = null;
@@ -463,7 +466,7 @@ class CaptureService {
       if (frame) {
         clog('click@', clickAt, 'frame', frame.source || 'loop',
           'started', frame.startedAt - clickAt, 'ms, captured', frame.capturedAt - clickAt, 'ms rel. click');
-        const result = this.storeFrameAsStep(guideId, frame.mode, frame, clickPos);
+        const result = await this.storeFrameAsStep(guideId, frame.mode, frame, clickPos, clickMeta);
         if (result.ok) this.noteStepAdded(result.step, trigger, guideId);
         return result;
       }
@@ -1211,6 +1214,9 @@ public static class SFMouseHook {
     let clickPos = osPoint ? this.osPointToDip(osPoint) : null;
     if (!clickPos) clickPos = this.screen.getCursorScreenPoint();
     clog('click@', clickAt, button, 'os', osPoint, '-> dip', clickPos);
+    this.pendingClickOsPoint = osPoint && Number.isFinite(osPoint.x) && Number.isFinite(osPoint.y)
+      ? { x: osPoint.x, y: osPoint.y }
+      : null;
     this.enqueueClickCapture(clickPos, clickAt, button || 'mouse');
   }
 
@@ -1268,7 +1274,17 @@ public static class SFMouseHook {
    * fast the user clicks or how slow the encoder is.
    */
   enqueueClickCapture(clickPos, clickAt = Date.now(), button = 'mouse') {
-    const clickMeta = { at: Number.isFinite(clickAt) ? clickAt : Date.now(), button: button || 'mouse' };
+    const osPoint = this.pendingClickOsPoint && Number.isFinite(this.pendingClickOsPoint.x) && Number.isFinite(this.pendingClickOsPoint.y)
+      ? this.pendingClickOsPoint
+      : null;
+    this.pendingClickOsPoint = null;
+    const clickMeta = {
+      at: Number.isFinite(clickAt) ? clickAt : Date.now(),
+      button: button || 'mouse',
+    };
+    if (osPoint) {
+      clickMeta.osPoint = { x: osPoint.x, y: osPoint.y };
+    }
     if (this.session && !this.session.paused && !this.userIsInApp()) {
       // The guide id pins the click to its recording so it can still be
       // stored if the session stops while this click waits in the queue.
@@ -1300,7 +1316,18 @@ public static class SFMouseHook {
     };
   }
 
-  storeFrameAsStep(guideId, mode, frame, clickPos = null) {
+  async buildStepTitle(mode, frame, clickPos = null, clickMeta = null) {
+    try {
+      if (this.textIntel && typeof this.textIntel.buildCaptureTitle === 'function') {
+        return await this.textIntel.buildCaptureTitle({ mode, frame, clickPos, clickMeta });
+      }
+    } catch {
+      // fall back to the local semantic title below
+    }
+    return this.autoTitle(mode);
+  }
+
+  async storeFrameAsStep(guideId, mode, frame, clickPos = null, clickMeta = null) {
     if (!frame) return { ok: false, reason: 'no capture frame available' };
     const annotations = [];
     // The click position (DIP, read at event time) wins over the frame's
@@ -1324,7 +1351,7 @@ public static class SFMouseHook {
     }
 
     const step = this.store.addStep(guideId, {
-      title: this.autoTitle(mode),
+      title: await this.buildStepTitle(mode, frame, clickPos, clickMeta),
       annotations,
       focusedView: {
         enabled: Boolean(this.settings.get('editor.focusedViewDefaultForNewSteps')),
@@ -1335,14 +1362,7 @@ public static class SFMouseHook {
   }
 
   autoTitle(mode) {
-    const tplStr = this.settings.get('editor.autoTitleTemplate') || '[[Mode]] capture [[Time]]';
-    const now = new Date();
-    const pad = (n) => String(n).padStart(2, '0');
-    return expandPlaceholders(tplStr, {
-      Mode: { fullscreen: 'Screen', window: 'Window', region: 'Region' }[mode] || 'Screen',
-      Time: `${pad(now.getHours())}:${pad(now.getMinutes())}:${pad(now.getSeconds())}`,
-      Date: `${now.getFullYear()}-${pad(now.getMonth() + 1)}-${pad(now.getDate())}`,
-    });
+    return DEFAULT_CAPTURE_TITLES[mode] || 'Capture';
   }
 
   /** Grab the screen/window image as { image, display } or throw. */
@@ -1451,8 +1471,12 @@ public static class SFMouseHook {
     const cropped = image.crop(rect);
     const size = cropped.getSize();
     if (!size.width || !size.height) return { ok: false, reason: 'empty selection' };
-    const step = this.store.addStep(guideId, { title: this.autoTitle('region') },
-      cropped.toPNG(), size);
+    const step = await this.storeFrameAsStep(guideId, 'region', {
+      image: cropped,
+      size,
+      display,
+      cursor: null,
+    }, null, null);
     return { ok: true, step };
   }
 

app/main.js

@@ -19,6 +19,7 @@ const { exportGuideArchive, importGuideArchive, saveLinkedGuide, readArchive } =
 const { createSnapshot, listSnapshots, restoreSnapshot } = require('../core/snapshots');
 const { readLock } = require('../core/locks');
 const CaptureService = require('./capture');
+const { TextIntelService } = require('./text-intel');
 const { keepProcessesResponsive } = require('./win-power');
 
 const APP_ID = 'com.stepforge.app';
@@ -59,6 +60,7 @@ let settings;
 let searchIndex;
 let templates;
 let capture;
+let textIntel;
 let mainWindow;
 
 function reindex(guideId) {
@@ -503,6 +505,22 @@ function setupIpc() {
     if (keyPath.startsWith('capture.hotkey')) registerHotkeys();
     return settings.data;
   });
+  h('ai:test', async ({ enabled = null, ollama = null } = {}) => {
+    return textIntel.testAiConnection({
+      enabled,
+      ollama,
+    });
+  });
+  h('ai:fillStep', async ({ guideId, stepId, target = 'all', blockId = null } = {}) => {
+    const result = await textIntel.generateStepPatch({
+      guideId,
+      stepId,
+      target,
+      blockId,
+    });
+    if (result.ok) reindex(guideId);
+    return result;
+  });
   h('placeholders:globals:get', () => settings.getGlobalPlaceholders());
   h('placeholders:globals:set', (values) => settings.setGlobalPlaceholders(values));
 
@@ -739,6 +757,13 @@ if (!gotLock) {
     settings = new Settings(store.settingsDir);
     searchIndex = new SearchIndex(store.indexDir);
     templates = new TemplateManager(store.templatesDir);
+    textIntel = new TextIntelService({
+      store,
+      settings,
+      getWindow: () => mainWindow,
+      dataDir,
+      screenApi: screen,
+    });
     // Bringing up the desktop-capture stream spawns/upgrades Chromium's GPU
     // and screen-capture utility processes — which can be born after a session
     // already started, so the start-time EcoQoS opt-out misses them. Re-apply
@@ -759,6 +784,7 @@ if (!gotLock) {
       settings,
       getWindow: () => mainWindow,
       notify: captureNotify,
+      textIntel,
     });
 
     applyTheme();
@@ -778,6 +804,9 @@ if (!gotLock) {
       capture.stopClickWatcher();
       capture.destroySessionTray();
     }
+    if (textIntel) {
+      textIntel.shutdown().catch(() => {});
+    }
     // clean preview temp files on close
     try {
       for (const entry of fs.readdirSync(store.tempDir)) {

app/preload.js

@@ -52,6 +52,10 @@ const api = {
     globalPlaceholders: invoke('placeholders:globals:get'),
     setGlobalPlaceholders: invoke('placeholders:globals:set'),
   },
+  ai: {
+    test: invoke('ai:test'),
+    fillStep: invoke('ai:fillStep'),
+  },
   capture: {
     shoot: invoke('capture:shoot'),
     region: invoke('capture:region'),

app/renderer/app.js

@@ -117,6 +117,7 @@ class StepForgeApp {
       guideFolders: library.folders?.guideFolders || {},
     };
     this.state.trash = trash;
+    this.editor.setSettings(settings);
   }
 
   async refreshLibrary({ keepFilter = true } = {}) {
@@ -320,6 +321,7 @@ class StepForgeApp {
             { label: 'Guide information…', action: () => this.editor.openGuideInfo() },
             { label: 'Guide placeholders…', action: () => this.editor.openGuidePlaceholders() },
             { label: 'Backups & snapshots…', action: () => this.editor.openBackupsDialog() },
+            { label: 'Generate all text fields with AI', action: () => this.editor.generateAllTextFieldsWithAi() },
             { label: guide && guide.linkedSource ? 'Linked guide…' : 'Linked guide (not linked)', action: () => this.editor.openLinkedGuide() },
             'sep',
             { label: 'Keyboard shortcuts…', action: () => this.editor.openShortcutsHelp() },
@@ -866,6 +868,7 @@ class StepForgeApp {
         await api.settings.set({ keyPath: 'spellcheck', value: next.spellcheck });
         await api.settings.set({ keyPath: 'capture', value: next.capture });
         await api.settings.set({ keyPath: 'editor', value: next.editor });
+        await api.settings.set({ keyPath: 'ai', value: next.ai });
         await api.settings.set({ keyPath: 'exports', value: next.exports });
         await api.settings.set({ keyPath: 'backups', value: next.backups });
         await api.settings.setGlobalPlaceholders(next.placeholders || {});

app/renderer/dialogs.js

@@ -313,6 +313,42 @@ function showSettingsDialog({
     const captureOutside = el('input', { type: 'checkbox', checked: Boolean(settings.capture?.captureOutsideClicks) });
     const confirmSimple = el('input', { type: 'checkbox', checked: Boolean(settings.capture?.confirmSimpleCapture) });
     const keepLast = makeInput(settings.backups?.keepLast ?? 10, 'number', { min: 0, step: 1 });
+    const aiEnabled = el('input', { type: 'checkbox', checked: Boolean(settings.ai?.enabled) });
+    const ollamaHost = makeInput(settings.ai?.ollama?.host || 'http://127.0.0.1:11434');
+    const ollamaModel = makeInput(settings.ai?.ollama?.model || 'llama3.2:1b');
+    const aiStatus = el('div', { className: 'muted ai-status' }, 'AI stays local through Ollama. Nothing is sent to the cloud.');
+    const testAiBtn = el('button', { type: 'button' }, 'Test connection');
+
+    const updateAiStatus = (message, { error = false } = {}) => {
+      aiStatus.textContent = message;
+      aiStatus.classList.toggle('error', Boolean(error));
+    };
+
+    const testAiConnection = async () => {
+      setButtonLoading(testAiBtn, true, 'Testing…');
+      updateAiStatus('Checking Ollama at the configured host…');
+      try {
+        const result = await api.ai.test({
+          ollama: {
+            host: ollamaHost.value.trim(),
+            model: ollamaModel.value.trim(),
+          },
+        });
+        if (!result.ok) {
+          updateAiStatus(result.reason || 'Could not connect to Ollama.', { error: true });
+          return;
+        }
+        if (result.installed) {
+          updateAiStatus(`Connected to ${result.host} with ${result.model}.`);
+        } else {
+          updateAiStatus(`Connected to ${result.host}. Model ${result.model} is not installed yet.`, { error: true });
+        }
+      } catch (err) {
+        updateAiStatus(err.message || 'Could not connect to Ollama.', { error: true });
+      } finally {
+        setButtonLoading(testAiBtn, false);
+      }
+    };
 
     const placeholderRows = el('div', { className: 'placeholder-rows' });
     const rows = [];
@@ -369,6 +405,19 @@ function showSettingsDialog({
         el('legend', {}, 'Backups'),
         labeledRow('Keep last snapshots', keepLast),
       ),
+      el('fieldset', {},
+        el('legend', {}, 'AI'),
+        labeledRow('Enable AI text filling', aiEnabled),
+        labeledRow('Ollama host', ollamaHost),
+        labeledRow('Ollama model', ollamaModel),
+        el('div.row', { style: { justifyContent: 'space-between' } },
+          aiStatus,
+          testAiBtn,
+        ),
+        el('div.muted', {},
+          'AI generation is manual only. Captures stay deterministic until you click a generate button.',
+        ),
+      ),
       el('fieldset', {},
         el('legend', {}, 'Global placeholders'),
         placeholderRows,
@@ -412,6 +461,15 @@ function showSettingsDialog({
                 ...settings.backups,
                 keepLast: Number(keepLast.value || 0),
               },
+              ai: {
+                ...settings.ai,
+                enabled: aiEnabled.checked,
+                ollama: {
+                  ...(settings.ai?.ollama || {}),
+                  host: ollamaHost.value.trim(),
+                  model: ollamaModel.value.trim(),
+                },
+              },
               placeholders: rows.reduce((acc, row) => {
                 const inputs = row.querySelectorAll('input');
                 const key = inputs[0].value.trim();
@@ -430,6 +488,14 @@ function showSettingsDialog({
     });
 
     form.addEventListener('submit', (e) => e.preventDefault());
+    testAiBtn.addEventListener('click', testAiConnection);
+    aiEnabled.addEventListener('change', () => {
+      updateAiStatus(
+        aiEnabled.checked
+          ? 'AI generation will be available once Ollama is reachable.'
+          : 'AI generation is disabled. The settings are still saved for later.',
+      );
+    });
   });
 }
 

app/renderer/editor.js

@@ -135,6 +135,7 @@ class GuideEditor {
     this.descriptionDirty = false;
     this.titleDirty = false;
     this.active = true;
+    this.settings = {};
 
     this.saveStepDebounced = debounce(() => this.flushStep(), 180);
     this.saveGuideDebounced = debounce(() => this.flushGuide(), 180);
@@ -152,6 +153,80 @@ class GuideEditor {
     this.active = Boolean(active);
   }
 
+  setSettings(settings) {
+    this.settings = settings || {};
+    this.updateAiButtonState();
+  }
+
+  isAiEnabled() {
+    return Boolean(this.settings?.ai?.enabled);
+  }
+
+  updateAiButtonState() {
+    const enabled = this.isAiEnabled();
+    const buttons = [
+      this.dom?.titleAiBtn,
+      this.dom?.descAiBtn,
+      ...(this.dom?.blocksList ? [...this.dom.blocksList.querySelectorAll('button[data-ai-action]')] : []),
+    ].filter(Boolean);
+    for (const button of buttons) {
+      button.disabled = !enabled;
+      button.title = enabled
+        ? button.dataset.aiTitle || 'Generate with AI'
+        : 'Enable AI in Settings first.';
+    }
+  }
+
+  async runAiGeneration(target, { blockId = null, button = null } = {}) {
+    if (!this.currentStep) {
+      this.onToast('Select a step first.', { error: true });
+      return null;
+    }
+    if (!this.isAiEnabled()) {
+      this.onToast('Enable AI in Settings first.', { error: true });
+      return null;
+    }
+    if (this.pendingSave) await this.flushStep();
+    if (button) setButtonLoading(button, true, 'Generating…');
+    try {
+      const result = await api.ai.fillStep({
+        guideId: this.guideId,
+        stepId: this.currentStep.stepId,
+        target,
+        blockId,
+      });
+      if (!result || !result.ok) {
+        this.onToast(result?.reason || 'AI generation failed.', { error: true });
+        return null;
+      }
+      await this.reload(result.step.stepId);
+      this.onToast('AI text filled.');
+      return result.step;
+    } catch (err) {
+      this.onToast(err.message || 'AI generation failed.', { error: true });
+      return null;
+    } finally {
+      if (button) setButtonLoading(button, false);
+    }
+  }
+
+  async generateTitleWithAi(button = null) {
+    return this.runAiGeneration('title', { button });
+  }
+
+  async generateDescriptionWithAi(button = null) {
+    return this.runAiGeneration('description', { button });
+  }
+
+  async generateAllTextFieldsWithAi(button = null) {
+    return this.runAiGeneration('all', { button });
+  }
+
+  async generateBlockWithAi(kind, block, button = null) {
+    if (!block) return null;
+    return this.runAiGeneration('block', { blockId: block.id, button });
+  }
+
   get currentStep() {
     return this.stepMap.get(this.selectedStepId) || null;
   }
@@ -271,7 +346,14 @@ class GuideEditor {
       el('aside.pane-props', {},
         el('section', {},
           el('h3', {}, 'Step'),
-          this.dom.titleInput = el('input', { type: 'text', placeholder: 'Step title' }),
+          el('div.row', {},
+            this.dom.titleInput = el('input', { type: 'text', placeholder: 'Step title', style: { flex: 1 } }),
+            this.dom.titleAiBtn = el('button.ai', {
+              type: 'button',
+              title: 'Generate the step title with AI',
+              dataset: { aiAction: 'title', aiTitle: 'Generate the step title with AI' },
+            }, 'AI'),
+          ),
           this.dom.statusSelect = makeSelect('todo', [
             { value: 'todo', label: 'Todo' },
             { value: 'in-progress', label: 'In progress' },
@@ -296,7 +378,14 @@ class GuideEditor {
           ),
         ),
         el('section', {},
-          el('h3', {}, 'Description'),
+          el('div.row', { style: { justifyContent: 'space-between', alignItems: 'center' } },
+            el('h3', { style: { margin: 0 } }, 'Description'),
+            this.dom.descAiBtn = el('button.ai', {
+              type: 'button',
+              title: 'Generate the description with AI',
+              dataset: { aiAction: 'description', aiTitle: 'Generate the description with AI' },
+            }, 'AI'),
+          ),
           this.dom.richToolbar = el('div.rich-toolbar', {},
             this.toolbarBtn('bold', 'Bold'),
             this.toolbarBtn('italic', 'Italic'),
@@ -416,6 +505,7 @@ class GuideEditor {
       this.renderStepList();
       this.emitMeta();
     });
+    this.dom.titleAiBtn.addEventListener('click', () => this.generateTitleWithAi(this.dom.titleAiBtn));
 
     this.dom.statusSelect.addEventListener('change', () => {
       if (!this.currentStep) return;
@@ -491,6 +581,7 @@ class GuideEditor {
     document.addEventListener('selectionchange', () => {
       if (document.activeElement === this.dom.descEditor) this.updateToolbarState();
     });
+    this.dom.descAiBtn.addEventListener('click', () => this.generateDescriptionWithAi(this.dom.descAiBtn));
 
     this.dom.descEditor.addEventListener('paste', (e) => {
       // Keep pasted text simple; backend sanitization will handle the rest.
@@ -504,6 +595,8 @@ class GuideEditor {
       if (!item) return;
       this.canvas.select(item.dataset.annId);
     });
+
+    this.updateAiButtonState();
   }
 
   renderAll() {
@@ -513,6 +606,7 @@ class GuideEditor {
     this.renderCanvas();
     this.renderAnnotationPanel();
     this.renderBlocksPanel();
+    this.updateAiButtonState();
     this.emitMeta();
   }
 
@@ -670,6 +764,15 @@ class GuideEditor {
         el('span.spacer'),
         el('button.icon', { type: 'button', title: 'Move block up', disabled: !canMoveUp, onClick: moveUp, dataset: { blockMove: 'up' } }, '↑'),
         el('button.icon', { type: 'button', title: 'Move block down', disabled: !canMoveDown, onClick: moveDown, dataset: { blockMove: 'down' } }, '↓'),
+        (() => {
+          const aiBtn = el('button.ai', {
+            type: 'button',
+            title: 'Generate this block with AI',
+            dataset: { aiAction: 'block', aiTitle: 'Generate this block with AI' },
+            onClick: () => this.generateBlockWithAi(kind, block, aiBtn),
+          }, 'AI');
+          return aiBtn;
+        })(),
         removeBtn(() => {
           if (kind === 'text') step.textBlocks = (step.textBlocks || []).filter((b) => b !== block);
           else if (kind === 'code') step.codeBlocks = (step.codeBlocks || []).filter((b) => b !== block);
@@ -761,6 +864,7 @@ class GuideEditor {
     if (!blocks.length) {
       this.dom.blocksList.append(el('div.muted', {}, 'Informational text, code, and table blocks can be reordered with drag handles or arrows.'));
     }
+    this.updateAiButtonState();
   }
 
   renderStepList() {
@@ -1658,6 +1762,7 @@ class GuideEditor {
         await api.settings.set({ keyPath: 'spellcheck', value: next.spellcheck });
         await api.settings.set({ keyPath: 'capture', value: next.capture });
         await api.settings.set({ keyPath: 'editor', value: next.editor });
+        await api.settings.set({ keyPath: 'ai', value: next.ai });
         await api.settings.set({ keyPath: 'exports', value: next.exports });
         await api.settings.set({ keyPath: 'backups', value: next.backups });
         await api.settings.setGlobalPlaceholders(next.placeholders || {});

app/renderer/style.css

@@ -90,6 +90,16 @@ button.tool.active {
   border-color: var(--accent);
   color: var(--accent-fg);
 }
+button.ai {
+  padding: 5px 10px;
+  border-color: color-mix(in srgb, var(--accent) 38%, var(--border));
+  background: color-mix(in srgb, var(--accent) 10%, var(--panel-solid));
+  color: var(--accent-strong);
+  font-weight: 650;
+  letter-spacing: 0.02em;
+}
+button.ai:hover { background: color-mix(in srgb, var(--accent) 16%, var(--panel-2)); }
+button.ai:disabled { opacity: 0.6; }
 button:disabled { opacity: 0.6; cursor: default; }
 button.loading {
   display: inline-flex;
@@ -751,6 +761,7 @@ fieldset legend {
   color: var(--muted);
   line-height: 1.5;
 }
+.ai-status.error { color: var(--danger); }
 
 #modal-root:not(:empty) {
   position: fixed;

app/text-intel.js

@@ -0,0 +1,438 @@
+'use strict';
+
+const fs = require('node:fs');
+const path = require('node:path');
+const { execFileSync } = require('node:child_process');
+
+const { createWorker } = require('tesseract.js');
+const {
+  buildCaptureTitle,
+  normalizeOllamaHost,
+  normalizeAiPatch,
+  buildAiPrompt,
+  applyAiPatchToStep,
+  displayText,
+  normalizeWhitespace,
+} = require('../core/text-intel');
+
+const OCR_CROP = {
+  width: 420,
+  height: 220,
+};
+
+function hasBinary(name) {
+  try {
+    execFileSync('which', [name], { stdio: 'pipe' });
+    return true;
+  } catch {
+    return false;
+  }
+}
+
+function clamp(v, min, max) {
+  return Math.min(max, Math.max(min, v));
+}
+
+class TextIntelService {
+  constructor({
+    store,
+    settings,
+    getWindow = () => null,
+    dataDir,
+    fetchImpl = global.fetch,
+    screenApi = null,
+  }) {
+    this.store = store;
+    this.settings = settings;
+    this.getWindow = getWindow;
+    this.dataDir = dataDir;
+    this.fetch = fetchImpl;
+    this.screen = screenApi;
+    this.worker = null;
+    this.workerPromise = null;
+    this.workerQueue = Promise.resolve();
+    this.ocrDataDir = path.join(dataDir, 'ocr', 'eng');
+  }
+
+  async shutdown() {
+    if (this.worker) {
+      try {
+        await this.worker.terminate();
+      } catch {
+        // best effort
+      }
+      this.worker = null;
+      this.workerPromise = null;
+    }
+  }
+
+  ensureLangData() {
+    const packageDir = path.dirname(require.resolve('@tesseract.js-data/eng/package.json'));
+    const source = path.join(packageDir, '4.0.0_best_int', 'eng.traineddata.gz');
+    const targetDir = this.ocrDataDir;
+    const target = path.join(targetDir, 'eng.traineddata.gz');
+    if (!fs.existsSync(target)) {
+      fs.mkdirSync(targetDir, { recursive: true });
+      fs.copyFileSync(source, target);
+    }
+    return targetDir;
+  }
+
+  async getWorker() {
+    if (this.workerPromise) return this.workerPromise;
+    this.workerPromise = (async () => {
+      const langPath = this.ensureLangData();
+      const worker = await createWorker('eng', 1, {
+        langPath,
+      });
+      await worker.setParameters({
+        preserve_interword_spaces: '1',
+      });
+      this.worker = worker;
+      return worker;
+    })();
+    return this.workerPromise;
+  }
+
+  async recognizeCrop(image, rect = null) {
+    const worker = await this.getWorker();
+    const cropped = rect ? image.crop(rect) : image;
+    const buffer = cropped.toPNG();
+    const result = await worker.recognize(buffer);
+    const text = String(result?.data?.text || '').trim();
+    return {
+      text,
+      confidence: Number.isFinite(result?.data?.confidence) ? result.data.confidence : null,
+      raw: result,
+    };
+  }
+
+  cropRectForPoint(frame, clickPos, { width = OCR_CROP.width, height = OCR_CROP.height } = {}) {
+    if (!frame || !frame.size) return null;
+    const bounds = frame.display.bounds || { x: 0, y: 0, width: frame.size.width, height: frame.size.height };
+    const scaleX = frame.size.width / bounds.width;
+    const scaleY = frame.size.height / bounds.height;
+    const point = clickPos || {
+      x: bounds.x + bounds.width / 2,
+      y: bounds.y + bounds.height / 2,
+    };
+    const centerX = (point.x - bounds.x) * scaleX;
+    const centerY = (point.y - bounds.y) * scaleY;
+    const rectW = Math.max(1, Math.round(width * scaleX));
+    const rectH = Math.max(1, Math.round(height * scaleY));
+    const rect = {
+      x: Math.round(centerX - rectW / 2),
+      y: Math.round(centerY - rectH / 2),
+      width: rectW,
+      height: rectH,
+    };
+    rect.x = clamp(rect.x, 0, Math.max(0, frame.size.width - rect.width));
+    rect.y = clamp(rect.y, 0, Math.max(0, frame.size.height - rect.height));
+    rect.width = clamp(rect.width, 1, frame.size.width);
+    rect.height = clamp(rect.height, 1, frame.size.height);
+    return rect;
+  }
+
+  async ocrAroundClick(frame, clickPos) {
+    if (!frame || !frame.image) return { text: '', confidence: null };
+    const rect = this.cropRectForPoint(frame, clickPos);
+    try {
+      return await this.recognizeCrop(frame.image, rect);
+    } catch {
+      return { text: '', confidence: null };
+    }
+  }
+
+  async collectForegroundWindowContext(osPoint = null) {
+    try {
+      if (process.platform === 'win32') return this.collectWindowsWindowContext(osPoint);
+      if (process.platform === 'darwin') return this.collectMacWindowContext();
+      if (process.platform === 'linux') return this.collectLinuxWindowContext();
+    } catch {
+      // best effort only
+    }
+    return { appName: '', windowTitle: '' };
+  }
+
+  collectWindowsWindowContext(osPoint = null) {
+    const hasPoint = osPoint && Number.isFinite(osPoint.x) && Number.isFinite(osPoint.y);
+    const clickX = hasPoint ? Number(osPoint.x) : 0;
+    const clickY = hasPoint ? Number(osPoint.y) : 0;
+    const script = `
+      $clickX = ${clickX};
+      $clickY = ${clickY};
+      $elementLabel = '';
+      $elementRole = '';
+      $elementClass = '';
+      $elementProcessId = 0;
+      if (${hasPoint ? '$true' : '$false'}) {
+        try {
+          Add-Type -AssemblyName UIAutomationClient,UIAutomationTypes,WindowsBase | Out-Null
+          $point = New-Object System.Windows.Point($clickX, $clickY);
+          $element = [System.Windows.Automation.AutomationElement]::FromPoint($point);
+          if ($element) {
+            $current = $element.Current;
+            $elementLabel = $current.Name;
+            $elementRole = $current.LocalizedControlType;
+            $elementClass = $current.ClassName;
+            $elementProcessId = $current.ProcessId;
+          }
+        } catch { }
+      }
+      Add-Type @"
+using System;
+using System.Runtime.InteropServices;
+using System.Text;
+public static class Win32 {
+  [DllImport("user32.dll")] public static extern IntPtr GetForegroundWindow();
+  [DllImport("user32.dll", CharSet = CharSet.Unicode)]
+  public static extern int GetWindowText(IntPtr hWnd, StringBuilder text, int count);
+  [DllImport("user32.dll")] public static extern uint GetWindowThreadProcessId(IntPtr hWnd, out uint processId);
+}
+"@;
+      $hWnd = [Win32]::GetForegroundWindow();
+      $sb = New-Object System.Text.StringBuilder 512;
+      [void][Win32]::GetWindowText($hWnd, $sb, $sb.Capacity);
+      $pid = 0;
+      [void][Win32]::GetWindowThreadProcessId($hWnd, [ref]$pid);
+      $proc = Get-Process -Id $pid -ErrorAction SilentlyContinue | Select-Object -First 1;
+      $out = [ordered]@{
+        appName = if ($proc) { $proc.ProcessName } else { '' };
+        windowTitle = $sb.ToString();
+        elementLabel = $elementLabel;
+        elementRole = $elementRole;
+        elementClass = $elementClass;
+        elementProcessId = $elementProcessId;
+        pid = $pid;
+      };
+      $out | ConvertTo-Json -Compress;
+    `;
+    const result = execFileSync('powershell.exe', ['-NoProfile', '-NonInteractive', '-ExecutionPolicy', 'Bypass', '-Command', script], {
+      encoding: 'utf8',
+      stdio: ['ignore', 'pipe', 'pipe'],
+      timeout: 1200,
+    }).trim();
+    return JSON.parse(result || '{}');
+  }
+
+  collectMacWindowContext() {
+    const script = `
+      set appName to ""
+      set windowTitle to ""
+      tell application "System Events"
+        try
+          set frontApp to first application process whose frontmost is true
+          set appName to name of frontApp
+          try
+            set windowTitle to name of front window of frontApp
+          end try
+        end try
+      end tell
+      return appName & linefeed & windowTitle
+    `;
+    const result = execFileSync('osascript', ['-e', script], {
+      encoding: 'utf8',
+      stdio: ['ignore', 'pipe', 'pipe'],
+      timeout: 1200,
+    }).trimEnd();
+    const [appName = '', windowTitle = ''] = result.split(/\r?\n/);
+    return { appName, windowTitle };
+  }
+
+  collectLinuxWindowContext() {
+    if (!hasBinary('xprop')) return { appName: '', windowTitle: '' };
+    const active = execFileSync('xprop', ['-root', '_NET_ACTIVE_WINDOW'], {
+      encoding: 'utf8',
+      stdio: ['ignore', 'pipe', 'pipe'],
+      timeout: 1200,
+    });
+    const activeMatch = active.match(/window id # (0x[0-9a-fA-F]+)/);
+    if (!activeMatch) return { appName: '', windowTitle: '' };
+    const winId = activeMatch[1];
+    const details = execFileSync('xprop', ['-id', winId, '_NET_WM_NAME', 'WM_NAME', 'WM_CLASS'], {
+      encoding: 'utf8',
+      stdio: ['ignore', 'pipe', 'pipe'],
+      timeout: 1200,
+    });
+    const titleMatch = details.match(/(?:_NET_WM_NAME\(UTF8_STRING\)|WM_NAME\(STRING\)|WM_NAME\(UTF8_STRING\)) = "([^"]*)"/);
+    const classMatch = details.match(/WM_CLASS\(STRING\) = "([^"]*)"(?:, "([^"]*)")?/);
+    return {
+      appName: classMatch ? (classMatch[2] || classMatch[1] || '') : '',
+      windowTitle: titleMatch ? titleMatch[1] : '',
+    };
+  }
+
+  async buildCaptureTitle({ mode, frame, clickPos, clickMeta = null }) {
+    const [metadata, ocr] = await Promise.all([
+      this.collectForegroundWindowContext(clickMeta?.osPoint || null),
+      this.ocrAroundClick(frame, clickPos),
+    ]);
+    return buildCaptureTitle({
+      mode,
+      metadata,
+      ocrText: ocr.text,
+    });
+  }
+
+  aiEnabled() {
+    return Boolean(this.settings.get('ai.enabled'));
+  }
+
+  aiConfig(override = null) {
+    const stored = this.settings.get('ai') || {};
+    const merged = override ? {
+      ...stored,
+      ...override,
+      ollama: {
+        ...(stored.ollama || {}),
+        ...(override.ollama || {}),
+      },
+    } : stored;
+    return {
+      ...merged,
+      enabled: override && Object.prototype.hasOwnProperty.call(override, 'enabled')
+        ? Boolean(override.enabled)
+        : Boolean(stored.enabled),
+      ollama: {
+        host: normalizeOllamaHost(merged.ollama?.host || ''),
+        model: normalizeWhitespace(merged.ollama?.model || ''),
+      },
+    };
+  }
+
+  async testAiConnection(override = null) {
+    const config = this.aiConfig(override);
+    if (!config.ollama.host) {
+      return { ok: false, reason: 'Set an Ollama host first.' };
+    }
+    const tagsUrl = new URL('/api/tags', `${config.ollama.host.replace(/\/+$/, '')}/`);
+    const res = await this.fetch(tagsUrl, { method: 'GET' });
+    if (!res.ok) {
+      return { ok: false, reason: `Ollama check failed (${res.status})` };
+    }
+    const data = await res.json();
+    const models = Array.isArray(data?.models) ? data.models.map((model) => model.name).filter(Boolean) : [];
+    const installed = config.ollama.model ? models.includes(config.ollama.model) : false;
+    return {
+      ok: true,
+      installed,
+      models,
+      host: config.ollama.host,
+      model: config.ollama.model,
+    };
+  }
+
+  async callOllama({ host, model, prompt, systemPrompt }) {
+    const url = new URL('/api/chat', `${host.replace(/\/+$/, '')}/`);
+    const response = await this.fetch(url, {
+      method: 'POST',
+      headers: { 'Content-Type': 'application/json' },
+      body: JSON.stringify({
+        model,
+        stream: false,
+        format: 'json',
+        messages: [
+          { role: 'system', content: systemPrompt },
+          { role: 'user', content: prompt },
+        ],
+        options: {
+          temperature: 0.2,
+        },
+      }),
+    });
+    if (!response.ok) {
+      throw new Error(`Ollama request failed (${response.status})`);
+    }
+    const payload = await response.json();
+    const content = payload?.message?.content;
+    if (typeof content !== 'string' || !content.trim()) {
+      throw new Error('Ollama returned an empty response');
+    }
+    return content;
+  }
+
+  async generateStepPatch({
+    guideId,
+    stepId,
+    target = 'all',
+    blockId = null,
+  }) {
+    try {
+      const config = this.aiConfig();
+      if (!config.enabled) {
+        return { ok: false, reason: 'Enable AI in settings first.' };
+      }
+      if (!config.ollama.host || !config.ollama.model) {
+        return { ok: false, reason: 'Configure Ollama host and model in Settings.' };
+      }
+
+      const guide = this.store.getGuide(guideId);
+      const step = this.store.getStep(guideId, stepId);
+      if (!guide || !step) {
+        return { ok: false, reason: 'Guide or step not found.' };
+      }
+
+      const currentBlock = blockId
+        ? [...(step.textBlocks || []), ...(step.codeBlocks || []), ...(step.tableBlocks || [])].find((b) => b.id === blockId) || null
+        : null;
+      if (blockId && target === 'block' && !currentBlock) {
+        return { ok: false, reason: 'Block not found.' };
+      }
+
+      let captureContext = null;
+      if (step.image) {
+        const imagePath = this.store.stepImagePath(guideId, stepId, 'working') || this.store.stepImagePath(guideId, stepId, 'original');
+        if (imagePath && fs.existsSync(imagePath)) {
+          const { nativeImage } = require('electron');
+          const image = nativeImage.createFromPath(imagePath);
+          if (!image.isEmpty()) {
+            const clickPoint = this.clickPointFromStep(step, image);
+            const [metadata, ocr] = await Promise.all([
+              this.collectForegroundWindowContext(),
+              this.ocrAroundClick({ image, size: image.getSize(), display: { bounds: { x: 0, y: 0, width: image.getSize().width, height: image.getSize().height } } }, clickPoint),
+            ]);
+            captureContext = {
+              ...metadata,
+              ocrText: ocr.text,
+              mode: step.kind === 'image' ? 'fullscreen' : 'content',
+            };
+          }
+        }
+      }
+
+      const { systemPrompt, prompt } = buildAiPrompt({
+        target,
+        guide,
+        step,
+        captureContext,
+        block: currentBlock,
+      });
+
+      const raw = await this.callOllama({
+        host: config.ollama.host,
+        model: config.ollama.model,
+        prompt,
+        systemPrompt,
+      });
+      const patch = normalizeAiPatch(raw);
+      const updated = applyAiPatchToStep(step, patch, { target, blockId });
+      const saved = this.store.saveStep(guideId, updated);
+      return { ok: true, step: saved, patch };
+    } catch (err) {
+      return { ok: false, reason: err && err.message ? err.message : 'AI generation failed.' };
+    }
+  }
+
+  clickPointFromStep(step, image = null) {
+    const marker = (step.annotations || []).find((ann) => ann.type === 'oval' && Number.isFinite(ann.x) && Number.isFinite(ann.y) && Number.isFinite(ann.w) && Number.isFinite(ann.h));
+    if (!marker) return null;
+    const size = image ? image.getSize() : step.image?.size || { width: 0, height: 0 };
+    if (!size.width || !size.height) return null;
+    return {
+      x: Math.round((marker.x + marker.w / 2) * size.width),
+      y: Math.round((marker.y + marker.h / 2) * size.height),
+    };
+  }
+}
+
+module.exports = { TextIntelService };