docker_add

.gitignore

@@ -1,3 +1,9 @@
+# Models (FastEmbed cache)
+.cache/
+models/
+*.bin
+*.onnx
+
 # Python
 __pycache__/
 *.py[cod]
@@ -7,7 +13,11 @@ __pycache__/
 venv/
 .venv/
 env/
+ENV/
 .env
+.env.local
+.env.production
+*.env
 *.egg-info/
 .eggs/
 *.egg

CHANGES_IMAGE_PREVIEW_FIX.md

@@ -1,217 +0,0 @@
-# Image Preview and Groq API Fix - Summary
-
-## Changes Made
-
-### 1. Image Preview UI Fix (Frontend)
-
-**Issue:** Image preview was showing as a large card above the input field, not matching the desired inline thumbnail appearance from the screenshot.
-
-**Solution:** 
-- Removed the large preview card that appeared above the input
-- Added a small **60px inline thumbnail** that appears next to the input controls
-- Thumbnail includes:
-  - Clickable to view full size in lightbox
-  - Small close button overlay (top-right)
-  - Proper border and styling matching the purple theme
-  - Uses `objectFit: 'cover'` for clean thumbnail appearance
-
-**Files Modified:**
-- `frontend/src/pages/Chat.jsx` (lines 687-793)
-
-**Visual Changes:**
-```
-Before: [Large preview card]
-        [Input field with buttons]
-
-After:  [Input field with buttons] [60px thumbnail] 
-```
-
----
-
-### 2. Image Display in Chat Messages
-
-**Issue:** When sending an image, it wasn't appearing in the user's chat bubble.
-
-**Solution:**
-- Added `multimodal_data` with `image_url` to the user message object
-- This stores the base64 preview URL for immediate display
-- The existing message rendering code (lines 483-521) already handles displaying images in chat bubbles
-
-**Files Modified:**
-- `frontend/src/pages/Chat.jsx` (lines 241-266)
-
-**Code Added:**
-```javascript
-multimodal_data: {
-    image_url: imagePreview // Store preview URL for display
-}
-```
-
----
-
-### 3. Groq API Image Processing Error Handling (Backend)
-
-**Issue:** Groq API image processing errors were causing the entire multimodal chat to fail.
-
-**Solution:**
-- Improved error handling in the multimodal chat endpoint
-- Now catches and logs image processing errors without breaking the chat flow
-- Provides fallback context when image analysis fails
-- Better error messages for debugging
-
-**Files Modified:**
-- `backend/api/chat.py` (lines 212-240)
-
-**Error Handling Flow:**
-```python
-try:
-    processor = create_multimodal_processor()
-    image_result = processor.process_image(str(temp_path))
-    
-    if image_result.get("success"):
-        # Use AI-generated description
-        multimodal_context += f"\n[IMAGE DESCRIPTION]: {image_desc}"
-    else:
-        # Fallback: mention the image was uploaded
-        logger.warning(f"Image analysis failed: {error}")
-        multimodal_context += f"\n[IMAGE]: User uploaded an image"
-except Exception as e:
-    # Graceful degradation
-    logger.error(f"Image processing exception: {e}")
-    multimodal_context += f"\n[IMAGE]: User uploaded an image"
-```
-
----
-
-## Testing Checklist
-
-### Image Preview
-- [ ] Upload an image using the image upload button
-- [ ] Verify small 60px thumbnail appears inline with input controls
-- [ ] Click thumbnail to view full size in lightbox
-- [ ] Click close button (X) on thumbnail to remove
-- [ ] Verify thumbnail disappears after sending message
-
-### Chat Message Display
-- [ ] Send a message with an image attached
-- [ ] Verify image appears in your chat bubble as a thumbnail
-- [ ] Verify image can be clicked to view full size
-- [ ] Verify text message appears below the image
-
-### Groq API Processing
-- [ ] Check backend logs when sending an image
-- [ ] Verify "Analyzing image" log appears
-- [ ] If Groq API works: Should see image description in reasoning
-- [ ] If Groq API fails: Should see warning but chat still works
-
----
-
-## Common Issues and Solutions
-
-### Issue: GROQ_API_KEY not found
-**Solution:** Create `.env` file in `backend/` directory:
-```
-GROQ_API_KEY=your_groq_api_key_here
-```
-
-### Issue: Image processing fails with "model not found"
-**Solution:** Groq's vision model is: `llama-3.2-90b-vision-preview`
-- This is already configured in `utils/groq_client.py`
-- Ensure your API key has access to vision models
-
-### Issue: Image doesn't appear after sending
-**Solution:** Check:
-1. Browser console for errors
-2. Network tab to verify image was uploaded to Supabase
-3. Backend logs for processing errors
-
----
-
-## Architecture Overview
-
-### Frontend Flow
-```
-1. User selects image → handleFileSelect()
-2. FileReader creates base64 preview → setImagePreview()
-3. Preview shows as inline thumbnail
-4. User sends message → handleSend()
-5. Image included in userMessage.multimodal_data
-6. API call: sendMultimodalMessage()
-7. Preview clears, message appears in chat
-```
-
-### Backend Flow
-```
-1. Receive multimodal request at /api/chat/multimodal
-2. Upload image to Supabase Storage → image_url
-3. Save temp copy for AI processing
-4. Groq Vision analyzes image → description
-5. Description added to multimodal_context
-6. Reasoning engine processes query + context
-7. Return answer + image_url to frontend
-8. Cleanup temp file
-```
-
----
-
-## Files Changed Summary
-
-### Frontend
-- `frontend/src/pages/Chat.jsx`
-  - Removed large preview card (removed ~50 lines)
-  - Added inline 60px thumbnail preview (+50 lines)
-  - Added multimodal_data to user message (+3 lines)
-
-### Backend
-- `backend/api/chat.py`
-  - Improved image processing error handling (+16 lines)
-  - Added try-catch for graceful degradation
-  
-- `backend/test_groq_vision.py` (new file)
-  - Diagnostic script to test Groq configuration
-
----
-
-## Next Steps
-
-1. **Test the changes:**
-   - Start backend: `cd backend && uvicorn main:app --reload`
-   - Frontend should already be running
-   - Upload and send an image
-
-2. **Verify Groq API:**
-   ```bash
-   cd backend
-   python test_groq_vision.py
-   ```
-
-3. **Check logs** if issues occur:
-   - Backend console for API errors
-   - Browser DevTools console for frontend errors
-   - Network tab for upload status
-
----
-
-## Visual Reference
-
-Based on your screenshot, the final result should look like:
-
-```
-┌─────────────────────────────────────────────────┐
-│ Input field text here...                        │
-│                                                  │
-│ [🎤] [📷] [60x60 img] [Send ➤]                  │
-│         thumbnail                               │
-└─────────────────────────────────────────────────┘
-```
-
-When sent, appears in chat as:
-```
-User bubble:
-┌────────────────┐
-│  [thumbnail]   │ ← clickable
-│                │
-│ Your message   │
-│ text here      │
-└────────────────┘
-```

COMPLETE_FIX_SUMMARY.md

@@ -1,316 +0,0 @@
-# Complete Fix Summary - Image Preview & Groq API
-
-## ✅ All Issues Fixed
-
-### Issue 1: Image Preview Position ✓
-**Problem:** Image was showing as inline thumbnail, not matching your reference screenshots
-
-**Solution:** Restored large preview card ABOVE the input field
-- Preview now appears above the input (like screenshot #3)
-- Max size: 300px wide, 200px tall
-- Close button in top-right corner
-- Click to view full-size in lightbox
-- Purple border matching app theme
-
-**Files Changed:**
-- `frontend/src/pages/Chat.jsx` (lines 691-744)
-
----
-
-### Issue 2: Duplicate Preview Removed ✓
-**Problem:** There were two previews (above AND inline)
-
-**Solution:** Removed the inline 60px thumbnail
-- Only one preview now - the large one above input
-- Cleaner UI matching your screenshots
-
-**Files Changed:**
-- `frontend/src/pages/Chat.jsx` (lines 785-853)
-
----
-
-### Issue 3: Groq API Not Recognizing Images ✓
-**Problem:** Groq was returning "I don't have information about the image"
-
-**Solution:** Added comprehensive logging to track the entire flow
-- Added logging at every step of image processing
-- File size validation
-- Base64 encoding verification
-- API call tracking
-- Detailed error messages
-
-**Files Changed:**
-- `backend/utils/groq_client.py` (lines 156-230)
-- `backend/api/chat.py` (lines 220-270)
-
-**Logging Format:**
-```
-[MULTIMODAL] Image uploaded to Supabase: https://...
-[MULTIMODAL] Saving temp file: data/temp/abc123.jpg
-[MULTIMODAL] Temp file saved, size: 45678 bytes
-[MULTIMODAL] Starting image analysis with Groq Vision...
-[GROQ VISION] Starting image analysis for: data/temp/abc123.jpg
-[GROQ VISION] Image file size: 45678 bytes
-[GROQ VISION] Image encoded to base64, length: 61234 chars
-[GROQ VISION] Detected MIME type: image/jpeg
-[GROQ VISION] Calling Groq API with model: llama-3.2-90b-vision-preview
-[GROQ VISION] Success! Response length: 234 chars
-[GROQ VISION] Response preview: This image shows a financial literacy infographic...
-[MULTIMODAL] ✓ Image analyzed successfully
-```
-
----
-
-## Testing Steps
-
-### 1. Test Image Preview (Frontend)
-
-1. **Navigate to any agent chat**
-2. **Click the image upload button** 📷
-3. **Select an image file**
-4. **Verify:**
-   - ✓ Large preview appears ABOVE the input field
-   - ✓ Preview is max 300x200px
-   - ✓ Close button (X) appears in top-right
-   - ✓ Click image to view full-size
-   - ✓ Click X to remove preview
-5. **Type a message** describing the image
-6. **Click Send**
-7. **Verify:**
-   - ✓ Preview disappears from input area
-   - ✓ Image appears in YOUR message bubble (right side, purple)
-   - ✓ Image is clickable for full view
-
-### 2. Test Groq Image Recognition (Backend)
-
-1. **Open backend terminal** to watch logs
-2. **Upload and send an image** with text "what this image about"
-3. **Check backend logs** for:
-   ```
-   [MULTIMODAL] Image uploaded to Supabase...
-   [MULTIMODAL] Starting image analysis with Groq Vision...
-   [GROQ VISION] Starting image analysis...
-   [GROQ VISION] Success! Response length: XXX chars
-   ```
-4. **Verify in chat:**
-   - ✓ MEXAR responds with actual description of the image
-   - ✓ NOT "I don't have information about the image"
-   - ✓ Response shows confidence score
-   - ✓ "Explain reasoning" button available
-
-### 3. What to Look For in Logs
-
-**✓ SUCCESS PATTERN:**
-```
-[MULTIMODAL] Image uploaded to Supabase: https://...
-[MULTIMODAL] Temp file saved, size: 45678 bytes
-[GROQ VISION] Image encoded to base64, length: 61234 chars
-[GROQ VISION] Calling Groq API with model: llama-3.2-90b-vision-preview
-[GROQ VISION] Success! Response length: 234 chars
-[MULTIMODAL] ✓ Image analyzed successfully
-```
-
-**❌ ERROR PATTERNS:**
-
-**Pattern 1 - Missing API Key:**
-```
-[GROQ VISION] API call failed: ValueError: GROQ_API_KEY not found
-```
-**Fix:** Add GROQ_API_KEY to backend/.env
-
-**Pattern 2 - File Not Found:**
-```
-[MULTIMODAL] Image processing exception: FileNotFoundError
-```
-**Fix:** Check Supabase storage permissions
-
-**Pattern 3 - API Error:**
-```
-[GROQ VISION] API call failed: HTTPError: 401 Unauthorized
-```
-**Fix:** Check API key is valid
-
-**Pattern 4 - Model Not Available:**
-```
-[GROQ VISION] API call failed: Model not found
-```
-**Fix:** Verify Groq account has vision access
-
----
-
-## Visual Comparison
-
-### BEFORE (Your Issue)
-```
-┌─────────────────────────────────────┐
-│ [User Message with Image]          │
-│ [Small inline thumbnail]            │
-│ "what this image about"             │
-└─────────────────────────────────────┘
-
-└─[MEXAR Response]──────────────────┐
-│ "I don't have information about   │
-│  the image 'download (1).jpg'..." │
-│                                    │
-│ 🔴 NOT WORKING - No recognition    │
-└────────────────────────────────────┘
-
-Input: [inline 60px thumbnail] [text]
-```
-
-### AFTER (Fixed)
-```
-┌─[Large Preview Above Input]───┐
-│  ┌─────────────────────┐  [X]  │
-│  │                     │       │
-│  │   [Image Preview]   │       │
-│  │   (300x200px)       │       │
-│  │                     │       │
-│  └─────────────────────┘       │
-└───────────────────────────────┘
-
-Input: [🎤] [📷] [text field] [Send]
-
-
-└─[User Message]────────────────────┐
-│  ┌────────────┐                   │
-│  │  [Image]   │ ← clickable       │
-│  └────────────┘                   │
-│  "what this image about"          │
-└───────────────────────────────────┘
-
-└─[MEXAR Response]──────────────────┐
-│ "This image shows a financial     │
-│  literacy infographic with a      │
-│  light bulb and text about..."    │
-│                                    │
-│ ✅ WORKING - Image recognized!     │
-│ Confidence: 85%  [Explain]        │
-└────────────────────────────────────┘
-```
-
----
-
-## Common Issues & Solutions
-
-### Issue: Preview not appearing
-**Check:**
-1. Browser console for errors
-2. Image file type (jpg, png, gif, webp only)
-3. File size (should be < 10MB)
-
-### Issue: "I don't have information about the image"
-**Debug:**
-1. Check backend logs for `[GROQ VISION]` messages
-2. Look for API errors or exceptions
-3. Verify GROQ_API_KEY is set
-4. Test API key with: `cd backend && python test_groq_vision.py`
-
-### Issue: Image disappears after sending
-**This is normal!** The preview should:
-- Disappear from input area after sending
-- Appear in your message bubble
-- Stay visible in chat history
-
-If it's not appearing in message bubble:
-1. Check browser console
-2. Verify response includes `image_url`
-3. Check Supabase storage upload succeeded
-
----
-
-## Architecture Flow
-
-### Upload → Display → Send → AI Process
-
-```
-1. User selects image
-   ↓
-2. FileReader creates base64 preview
-   ↓
-3. Preview shows ABOVE input (300x200px)
-   ↓
-4. User types message + clicks Send
-   ↓
-5. Frontend: sendMultimodalMessage()
-   - Uploads original file to Supabase
-   - Includes base64 in message for display
-   ↓
-6. Backend: /api/chat/multimodal
-   - Saves temp copy of image
-   - Calls Groq Vision API
-   - Gets AI description
-   ↓
-7. Groq Vision: describe_image()
-   - Encodes to base64
-   - Sends to llama-3.2-90b-vision-preview
-   - Returns description
-   ↓
-8. Backend: Reasoning Engine
-   - Combines: user text + image description
-   - Generates answer
-   ↓
-9. Response to frontend
-   - Answer text
-   - Confidence score
-   - Image URL for display
-   - Explainability data
-   ↓
-10. Display in chat
-    - User bubble: image + text
-    - AI bubble: answer + confidence
-```
-
----
-
-## Files Modified Summary
-
-### Frontend (`frontend/src/pages/Chat.jsx`)
-- **Added:** Large preview card above input (lines 691-744)
-- **Removed:** Inline 60px thumbnail (lines 785-853)
-- **Result:** Single, large preview matching your screenshots
-
-### Backend (`backend/api/chat.py`)
-- **Enhanced:** Image processing logging (lines 220-270)
-- **Added:** Detailed step-by-step tracking
-- **Added:** Error type logging
-- **Result:** Full visibility into image processing
-
-### Backend (`backend/utils/groq_client.py`)
-- **Enhanced:** describe_image() function (lines 156-230)
-- **Added:** File validation
-- **Added:** API call logging
-- **Added:** Response preview logging
-- **Result:** Complete Groq API debugging
-
----
-
-## Next Steps
-
-1. **Test the changes** - Upload an image and verify:
-   - Preview appears above input (large, not inline)
-   - MEXAR recognizes and describes the image
-   - Backend logs show successful Groq API calls
-
-2. **Watch backend logs** - Look for:
-   - `[MULTIMODAL]` tags for upload/processing
-   - `[GROQ VISION]` tags for API calls
-   - Success messages with description preview
-
-3. **If Groq still fails:**
-   - Share the backend log output
-   - Check if GROQ_API_KEY has vision access
-   - Try test script: `python backend/test_groq_vision.py`
-
----
-
-## Success Criteria ✅
-
-- [ ] Image preview appears ABOVE input (like screenshot #3)
-- [ ] Preview is large (300x200px max), not tiny (60px)
-- [ ] Image shows in your message bubble after sending
-- [ ] MEXAR actually describes the image content
-- [ ] Backend logs show `[GROQ VISION] Success!`
-- [ ] No more "I don't have information about the image"
-
-All changes are complete and ready for testing!

FIX_SQLALCHEMY_F405.md

@@ -1,78 +0,0 @@
-# SQLAlchemy f405 Error Fix
-
-## Error
-```
-sqlalchemy.exc.AmbiguousForeignKeysError: 
-Could not determine join condition between parent/child tables on relationship
-(Background on this error at: https://sqlalche.me/e/20/f405)
-```
-
-## Root Cause
-The `User` model was missing the `conversations` relationship, causing SQLAlchemy to be unable to properly join the tables when querying conversations.
-
-## Fix Applied
-
-**File:** `backend/models/user.py`
-
-**Added:**
-```python
-from sqlalchemy.orm import relationship
-
-# Inside User class:
-# Relationships
-conversations = relationship("Conversation", backref="user", cascade="all, delete-orphan")
-```
-
-## Why This Fixes It
-
-The SQLAlchemy f405 error occurs when there's an ambiguous or missing relationship definition. In this case:
-
-- `Conversation` model had:
-  - `user_id = ForeignKey("users.id")`
-  - Trying to create a back-reference to User
-
-- `User` model was missing the corresponding relationship definition
-
-- SQLAlchemy couldn't determine how to join the tables
-
-By adding the `conversations` relationship to the User model:
-- ✅ Complete bidirectional relationship established
-- ✅ SQLAlchemy can now properly join User ↔ Conversation
-- ✅ Cascade delete works properly (when user deleted, conversations deleted)
-- ✅ No ambiguity in foreign key relationships
-
-## Testing
-
-The uvicorn server should have automatically reloaded. Try:
-
-1. Upload an image in the chat
-2. Send a message
-3. Check that no SQLAlchemy errors appear in backend logs
-4. Verify message is saved to database
-
-## Related Models
-
-All relationships are now properly defined:
-
-```
-User
- ↓ (one-to-many)
- conversations[] ✅
- agents[] ✅
-
-Agent
- ↓ (one-to-many)
- conversations[] ✅
- compilation_jobs[] ✅
- chunks[] ✅
-
-Conversation
- ↓ (one-to-many)
- messages[] ✅
- ↑ (many-to-one)
- user ✅ (via backref)
- agent ✅ (via back_populates)
-```
-
-## Status
-✅ **FIXED** - The relationship is now complete and the error should be resolved.

backend/Dockerfile

@@ -0,0 +1,28 @@
+FROM python:3.9-slim
+
+WORKDIR /app
+
+# Install system dependencies
+RUN apt-get update && apt-get install -y \
+    gcc \
+    g++ \
+    postgresql-client \
+    && rm -rf /var/lib/apt/lists/*
+
+# Copy requirements and install Python packages
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+# Copy application code
+COPY . .
+
+# Set environment for model caching to /tmp (only writable dir in HF Spaces)
+ENV HF_HOME=/tmp/.cache/huggingface
+ENV FASTEMBED_CACHE_PATH=/tmp/.cache/fastembed
+ENV SENTENCE_TRANSFORMERS_HOME=/tmp/.cache/sentence-transformers
+
+# Expose port 7860 (required by Hugging Face Spaces)
+EXPOSE 7860
+
+# Run FastAPI with uvicorn
+CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "7860"]

backend/main.py

@@ -73,9 +73,21 @@ app = FastAPI(
 )
 
 # Configure CORS
+# Configure CORS
+# CRITICAL: Configure CORS for Vercel frontend
+FRONTEND_URL = os.getenv("FRONTEND_URL", "http://localhost:3000")
+
+allow_origins = [
+    "*", 
+    FRONTEND_URL, 
+    "https://*.vercel.app", 
+    "http://localhost:3000",
+    "http://localhost:3001"
+]
+
 app.add_middleware(
     CORSMiddleware,
-    allow_origins=["*"],
+    allow_origins=allow_origins,
     allow_credentials=True,
     allow_methods=["*"],
     allow_headers=["*"],

backend/modules/knowledge_compiler.py

@@ -48,8 +48,13 @@ class KnowledgeCompiler:
         
         # Initialize embedding model (384 dim default)
         try:
-            self.embedding_model = TextEmbedding(model_name="BAAI/bge-small-en-v1.5")
-            logger.info("FastEmbed model loaded")
+            # Force cache to /tmp for HF Spaces or use env var
+            cache_dir = os.getenv("FASTEMBED_CACHE_PATH", "/tmp/.cache/fastembed")
+            self.embedding_model = TextEmbedding(
+                model_name="BAAI/bge-small-en-v1.5", 
+                cache_dir=cache_dir
+            )
+            logger.info(f"FastEmbed model loaded (cache: {cache_dir})")
         except Exception as e:
             logger.warning(f"Failed to load embedding model: {e}")
             self.embedding_model = None