feat: one-click Ollama installer, OS-aware download buttons, and AI parsing fixes

apps/extension-chrome/public/manifest.json

@@ -1,7 +1,7 @@
 {
   "manifest_version": 3,
   "name": "Offlyn Apply",
-  "version": "0.6.1",
+  "version": "0.6.3",
   "description": "Smart job application assistant - auto-fill forms, track applications, generate cover letters",
 
   "icons": {

apps/extension-chrome/public/onboarding/onboarding.html

@@ -1968,7 +1968,7 @@ <h2 class="step-title">Upload Your Resume</h2>
         <div class="content-card">
           <h2 class="step-title">Supercharge Your Applications with Local AI</h2>
           <p class="step-subtitle">Offlyn uses on-device AI — your data never leaves your computer.</p>
-          <p style="font-size:13px;color:#475569;margin:0 0 16px;line-height:1.6;">We believe AI should work for you privately. Offlyn runs AI models directly on your device so your resume, work history, and personal info stay 100% yours.</p>
+          <p style="font-size:13px;color:#475569;margin:0 0 16px;line-height:1.6;">We believe AI should work for you privately. Offlyn runs AI models directly on your device so your resume, work history, and personal info stay 100% yours. Once downloaded, the model runs fully offline — no data is ever sent to a server and there is no per-query cost.</p>
 
           <!-- State: Checking -->
           <div id="ollamaChecking" class="ollama-state active">
@@ -2084,6 +2084,7 @@ <h3 style="font-size:14px;font-weight:700;color:#1e2a3a;margin-bottom:6px;">Step
                   <svg width="14" height="14" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.5" stroke-linecap="round" stroke-linejoin="round"><polyline points="23 4 23 10 17 10"/><path d="M20.49 15a9 9 0 1 1-2.12-9.36L23 10"/></svg>
                   I Installed It — Check Again
                 </button>
+                <p id="helperErrorHint" style="font-size:11px;color:#dc2626;margin-top:8px;word-break:break-word;"></p>
               </div>
             </div>
 

apps/extension-chrome/src/background.ts

@@ -94,14 +94,18 @@ browser.runtime.onMessage.addListener(async (message: unknown, sender: browser.r
           const port = browser.runtime.connectNative(NATIVE_HOST_ID);
           port.postMessage({ cmd: 'ping' });
           port.onMessage.addListener((res: any) => {
+            console.log('[NativeMsg] ping response:', res);
             resolve({ installed: true, version: res.version ?? '?' });
             port.disconnect();
           });
           port.onDisconnect.addListener(() => {
-            resolve({ installed: false, error: (browser.runtime.lastError as any)?.message });
+            const errMsg = (browser.runtime.lastError as any)?.message ?? 'unknown disconnect';
+            console.error('[NativeMsg] connectNative disconnected:', errMsg);
+            resolve({ installed: false, error: errMsg });
           });
-        } catch {
-          resolve({ installed: false });
+        } catch (err) {
+          console.error('[NativeMsg] connectNative threw:', err);
+          resolve({ installed: false, error: String(err) });
         }
       });
     }
@@ -184,7 +188,9 @@ browser.runtime.onMessage.addListener(async (message: unknown, sender: browser.r
         if (legacyResult.status === 'rejected') warn('Legacy parser failed:', legacyResult.reason);
 
         if (!ragProfile && !legacyProfile) {
-          throw new Error('Both parsers failed — no profile could be extracted');
+          const ragErr    = ragResult.status    === 'rejected' ? String(ragResult.reason)    : 'no result';
+          const legacyErr = legacyResult.status === 'rejected' ? String(legacyResult.reason) : 'no result';
+          throw new Error(`Both parsers failed — RAG: ${ragErr} | Legacy: ${legacyErr}`);
         }
 
         let profile: any;

apps/extension-chrome/src/content.ts

@@ -267,7 +267,7 @@ function detectPage(): void {
     console.log('[OA] Page state:', classification.state, '|', classification.detectionReason);
 
     if (classification.state === 'NOT_JOB_PAGE') {
-      console.warn('[OA] Not a job page — skipping');
+      console.log('[OA] Not a job page — skipping');
       return;
     }
 

apps/extension-chrome/src/onboarding/onboarding.ts

@@ -422,7 +422,7 @@ async function checkOllamaConnection(): Promise<void> {
       }
     }
   } else {
-    console.warn('[Ollama Setup] Connection failed:', result.error);
+    console.log('[Ollama Setup] Connection failed:', result.error);
     showOllamaUIState('not-installed');
     // Determine whether the native helper is already installed and update sub-state
     checkNativeHelper().then(installed => updateHelperSubstate(installed));
@@ -432,9 +432,11 @@ async function checkOllamaConnection(): Promise<void> {
 // ── Native Messaging helpers ──────────────────────────────────────────────
 
 const HELPER_INSTALL_BASE =
-  'https://raw.githubusercontent.com/joelnishanth/offlyn-apply/main/scripts/native-host';
+  'https://raw.githubusercontent.com/rahulraonatarajan/offlyn-apply/Windows-ollama-setup/scripts/native-host';
 const HELPER_PKG_URL =
-  'https://github.com/joelnishanth/offlyn-apply/releases/download/v0.5.0/offlyn-helper.pkg';
+  'https://raw.githubusercontent.com/rahulraonatarajan/offlyn-apply/Windows-ollama-setup/scripts/native-host/install-mac-linux.sh';
+const HELPER_WIN_BAT_URL =
+  'https://raw.githubusercontent.com/rahulraonatarajan/offlyn-apply/Windows-ollama-setup/scripts/native-host/install-win.bat';
 
 function detectOS(): 'mac' | 'windows' | 'linux' {
   const ua = navigator.userAgent.toLowerCase();
@@ -474,7 +476,18 @@ function populateHelperInstructions(): void {
          class="btn btn-primary"
          style="display:inline-flex;align-items:center;gap:8px;font-size:14px;padding:10px 20px;text-decoration:none;margin-bottom:14px;">
         ${DOWNLOAD_SVG}
-        Download Installer
+        Download for Mac
+      </a>
+    `;
+  } else if (os === 'windows') {
+    container.innerHTML = `
+      <p style="font-size:13px;color:#475569;margin-bottom:16px;">Download and double-click the installer — it takes about 10 seconds. No terminal required.</p>
+      <a href="${HELPER_WIN_BAT_URL}"
+         target="_blank"
+         class="btn btn-primary"
+         style="display:inline-flex;align-items:center;gap:8px;font-size:14px;padding:10px 20px;text-decoration:none;margin-bottom:14px;">
+        ${DOWNLOAD_SVG}
+        Download for Windows (.bat)
       </a>
     `;
   } else {
@@ -493,8 +506,17 @@ function populateHelperInstructions(): void {
 async function checkNativeHelper(): Promise<boolean> {
   try {
     const res = await browser.runtime.sendMessage({ kind: 'CHECK_NATIVE_HELPER' });
-    return (res as any)?.installed === true;
-  } catch {
+    const installed = (res as any)?.installed === true;
+    if (!installed) {
+      const errMsg = (res as any)?.error ?? 'no error detail';
+      console.warn('[NativeHelper] not detected, reason:', errMsg);
+      // Show error in the helper status area so it is visible without DevTools
+      const hint = document.getElementById('helperErrorHint');
+      if (hint) hint.textContent = `Error: ${errMsg}`;
+    }
+    return installed;
+  } catch (err) {
+    console.error('[NativeHelper] sendMessage threw:', err);
     return false;
   }
 }
@@ -606,7 +628,21 @@ function setupOllamaStepListeners(): void {
                class="btn btn-primary"
                style="display:inline-flex;align-items:center;gap:8px;font-size:13px;padding:9px 16px;text-decoration:none;margin-bottom:10px;">
               ${DOWNLOAD_SVG}
-              Download Installer
+              Download for Mac
+            </a>
+            <p style="font-size:12px;color:#78350f;">After the installer finishes, click <strong>Re-test Connection</strong> below.</p>
+          `;
+        } else if (os === 'windows') {
+          corsWrap.innerHTML = `
+            <p style="font-size:13px;color:#92400e;margin-bottom:10px;">
+              <strong>To fix CORS automatically</strong>, install the Offlyn Helper first — it configures Ollama permissions in one step.
+            </p>
+            <a href="${HELPER_WIN_BAT_URL}"
+               target="_blank"
+               class="btn btn-primary"
+               style="display:inline-flex;align-items:center;gap:8px;font-size:13px;padding:9px 16px;text-decoration:none;margin-bottom:10px;">
+              ${DOWNLOAD_SVG}
+              Download for Windows (.bat)
             </a>
             <p style="font-size:12px;color:#78350f;">After the installer finishes, click <strong>Re-test Connection</strong> below.</p>
           `;

apps/extension-chrome/src/shared/error-classify.ts

@@ -31,8 +31,8 @@ export function enrichParseErrorMessage(raw: string): string {
   if (/not found|404|model .* not found/i.test(raw)) {
     return (
       'Required AI model not found. Please install the models by running:\n' +
-      'ollama pull llama3.2\n' +
-      'ollama pull nomic-embed-text'
+    'ollama pull llama3.2:1b\n' +
+    'ollama pull nomic-embed-text'
     );
   }
   return raw;

apps/extension-chrome/src/shared/mastra-agent.ts

@@ -4,8 +4,6 @@
  */
 
 import browser from './browser-compat';
-import { generateText } from 'ai';
-import { ollama } from 'ollama-ai-provider-v2';
 import type { Message } from './types';
 
 /**
@@ -41,19 +39,45 @@ class MastraAgentService {
   private baseUrl: string;
   private model: string;
   private embeddingModel: string;
-  private ollamaProvider: ReturnType<typeof ollama>;
   /** Set to false after the first embedding failure so callers can skip embedding-based retrieval */
   embeddingsAvailable = true;
 
-  constructor(baseUrl = 'http://localhost:11434', model = 'llama3.2') {
+  constructor(baseUrl = 'http://localhost:11434', model = 'llama3.2:1b') {
     this.baseUrl = baseUrl;
     this.model = model;
     this.embeddingModel = 'nomic-embed-text';
+  }
 
-    // Initialize Ollama provider with AI SDK (browser-compatible)
-    this.ollamaProvider = ollama(this.model, {
-      baseURL: this.baseUrl,
+  /**
+   * Direct fetch to Ollama /api/generate — avoids AI SDK version incompatibilities.
+   */
+  private async ollamaGenerate(
+    system: string,
+    prompt: string,
+    options: { temperature?: number; maxTokens?: number } = {}
+  ): Promise<string> {
+    const response = await fetch(`${this.baseUrl}/api/generate`, {
+      method: 'POST',
+      headers: { 'Content-Type': 'application/json' },
+      body: JSON.stringify({
+        model: this.model,
+        system,
+        prompt,
+        stream: false,
+        options: {
+          temperature: options.temperature ?? 0.1,
+          num_predict: options.maxTokens ?? 4096,
+        },
+      }),
     });
+
+    if (!response.ok) {
+      const errText = await response.text().catch(() => response.statusText);
+      throw new Error(`Ollama /api/generate returned ${response.status}: ${errText}`);
+    }
+
+    const data = await response.json();
+    return (data.response as string) ?? '';
   }
 
   /**
@@ -102,23 +126,15 @@ class MastraAgentService {
         messageCount: messages.length,
       });
 
-      // Convert messages to AI SDK format
       const systemMessage = messages.find((m) => m.role === 'system');
       const userMessages = messages.filter((m) => m.role !== 'system');
-
-      // Combine all user/assistant messages into a prompt
       const prompt = userMessages.map((msg) => msg.content).join('\n\n');
 
-      // Generate text using AI SDK — default 4096 to avoid truncating long JSON responses
-      const result = await generateText({
-        model: this.ollamaProvider,
-        system: systemMessage?.content,
-        prompt: prompt,
-        temperature: options?.temperature ?? 0.1,
-        maxTokens: options?.maxTokens ?? 4096,
-      });
-
-      const content = result.text;
+      const content = await this.ollamaGenerate(
+        systemMessage?.content ?? '',
+        prompt,
+        { temperature: options?.temperature, maxTokens: options?.maxTokens }
+      );
 
       if (!content) {
         throw new Error('No response content from AI agent');
@@ -229,17 +245,11 @@ Be thorough and capture ALL details. Return ONLY valid JSON with no markdown for
 
     const userPrompt = `${prompts[sectionType]}\n\nResume text:\n${sectionText}\n\nReturn ONLY the JSON (no markdown, no explanations):`;
 
-    const response = await generateText({
-      model: this.ollamaProvider,
-      system: systemPrompt,
-      prompt: userPrompt,
-      temperature: 0.1,
-      maxTokens: 3000,
-    });
+    const rawText = await this.ollamaGenerate(systemPrompt, userPrompt, { temperature: 0.1, maxTokens: 3000 });
 
     // Extract JSON - handle various response formats including malformed LLM output
-    const repaired = repairJSON(response.text);
-    if (repaired !== null && repaired !== undefined && !(Array.isArray(repaired) && (repaired as unknown[]).length === 0 && response.text.trim().length > 10)) {
+    const repaired = repairJSON(rawText);
+    if (repaired !== null && repaired !== undefined && !(Array.isArray(repaired) && (repaired as unknown[]).length === 0 && rawText.trim().length > 10)) {
       return repaired;
     }
     console.warn(`[AIAgent] No valid JSON found for ${sectionType}`);
@@ -375,14 +385,12 @@ Be thorough and capture ALL details. Return ONLY valid JSON with no markdown for
     // Generate summary
     onProgress?.('Generating summary...', 95);
     try {
-      const summaryResult = await generateText({
-        model: this.ollamaProvider,
-        system: 'Create a brief professional summary (2-3 sentences) from this resume text.',
-        prompt: chunks[0],
-        temperature: 0.3,
-        maxTokens: 200,
-      });
-      profile.summary = summaryResult.text.trim();
+      const summaryText = await this.ollamaGenerate(
+        'Create a brief professional summary (2-3 sentences) from this resume text.',
+        chunks[0],
+        { temperature: 0.3, maxTokens: 200 }
+      );
+      profile.summary = summaryText.trim();
     } catch {
       profile.summary = 'Professional with diverse experience and skills.';
     }
@@ -449,23 +457,17 @@ ${resumeText}
 
 Return ONLY the JSON object, nothing else:`;
 
-    const response = await generateText({
-      model: this.ollamaProvider,
-      system: systemPrompt,
-      prompt: userPrompt,
-      temperature: 0.1,
-      maxTokens: 4000,
-    });
+    const rawText = await this.ollamaGenerate(systemPrompt, userPrompt, { temperature: 0.1, maxTokens: 4000 });
 
-    console.log('[AIAgent] Raw response:', response.text.substring(0, 500));
+    console.log('[AIAgent] Raw response:', rawText.substring(0, 500));
 
     // Extract JSON from response
-    const repaired = repairJSON(response.text);
+    const repaired = repairJSON(rawText);
     if (repaired !== null && repaired !== undefined) {
       console.log('[AIAgent] Successfully parsed JSON');
       return repaired;
     }
-    console.error('[AIAgent] JSON repair failed for response:', response.text.substring(0, 200));
+    console.error('[AIAgent] JSON repair failed for response:', rawText.substring(0, 200));
     throw new Error('Could not parse JSON from response after repair attempt');
   }
 }

apps/extension-chrome/src/shared/ollama-client.ts

@@ -28,7 +28,7 @@ export interface ChatCompletionResponse {
 
 export class OllamaClient {
   private baseUrl = 'http://localhost:11434';
-  private model = 'llama3.2';
+  private model = 'llama3.2:1b';
   private embeddingModel = 'nomic-embed-text'; // Optimized for embeddings
 
   constructor(baseUrl?: string, model?: string) {
@@ -41,12 +41,11 @@ export class OllamaClient {
   private async loadStoredConfig(): Promise<void> {
     try {
       const config = await getOllamaConfig();
-      if (config.enabled) {
-        this.baseUrl = config.endpoint;
-        this.model = config.chatModel;
-        this.embeddingModel = config.embeddingModel;
-        console.log('[Ollama] Loaded config from storage:', config.endpoint, config.chatModel);
-      }
+      // Always load model names from stored config regardless of enabled state
+      this.baseUrl = config.endpoint;
+      this.model = config.chatModel;
+      this.embeddingModel = config.embeddingModel;
+      console.log('[Ollama] Loaded config from storage:', config.endpoint, config.chatModel);
     } catch (err) {
       console.warn('[Ollama] Failed to load stored config, using defaults:', err);
     }

apps/extension-chrome/src/shared/ollama-config.ts

@@ -15,7 +15,7 @@ export interface OllamaConfig {
 
 export const DEFAULT_OLLAMA_CONFIG: OllamaConfig = {
   endpoint: 'http://localhost:11434',
-  chatModel: 'llama3.2',
+  chatModel: 'llama3.2:1b',
   embeddingModel: 'nomic-embed-text',
   lastChecked: 0,
   enabled: false,

apps/extension-chrome/src/shared/ollama-service.ts

@@ -39,7 +39,7 @@ export async function checkOllamaConnection(): Promise<boolean> {
  */
 export async function analyzeFieldsWithOllama(
   prompt: string,
-  model: string = 'llama3.2'
+  model: string = 'llama3.2:1b'
 ): Promise<OllamaResponse | null> {
   try {
     const response = await fetch(`${OLLAMA_BASE_URL}/api/generate`, {
@@ -284,7 +284,7 @@ Respond with ONLY the value, nothing else. No quotes, no explanation, no preambl
       method: 'POST',
       headers: { 'Content-Type': 'application/json' },
       body: JSON.stringify({
-        model: 'llama3.2',
+        model: 'llama3.2:1b',
         prompt,
         stream: useStreaming
       })