# Shopify AI - Planning and Container Logs Fixes

## Summary
This document outlines the fixes applied to resolve issues with Mistral/Groq planning and container logs visibility.

## Issues Fixed

### 1. Groq Planning Not Working
**Problem:** When Groq was selected as the planning provider in the admin panel, no response was returned.

**Root Cause:** 
- Incorrect API endpoint URL: `https://api.groq.ai/v1` (wrong domain)
- Incorrect API request format (not using OpenAI-compatible format)
- Wrong response parsing logic

**Solution:**
- ✅ Updated Groq API URL to `https://api.groq.com/openai/v1/chat/completions`
- ✅ Changed request format to OpenAI-compatible (model + messages in payload)
- ✅ Fixed response parsing to extract from `data.choices[0].message.content`
- ✅ Added comprehensive logging for debugging
- ✅ Set default model to `llama-3.3-70b-versatile`
- ✅ Added model chain with fallback models

### 2. Mistral Planning Issues
**Problem:** Mistral planning might not work properly due to missing model information.

**Root Cause:** 
- The `sendMistralChat` function was not returning the model name in the response

**Solution:**
- ✅ Added `model` field to Mistral API response
- ✅ Ensured model information is tracked throughout the planning flow

### 3. Container Logs Not Visible
**Problem:** Even though extensive logging was added to the application, users couldn't see logs when using `docker logs`.

**Root Cause:** 
- In `scripts/entrypoint.sh` line 187, the Node.js server output was redirected to a file:
  ```bash
  node "$CHAT_APP_DIR/server.js" >/var/log/chat-service.log 2>&1 &
  ```
- This meant all `console.log()` and `console.error()` output was going to `/var/log/chat-service.log` inside the container
- Docker couldn't capture these logs because they weren't going to stdout/stderr

**Solution:**
- ✅ Removed the file redirection from entrypoint.sh
- ✅ Logs now go directly to stdout/stderr where Docker can capture them
- ✅ Users can now see all application logs using `docker logs <container_name>`

### 4. Missing Model Chain Support for Google and NVIDIA
**Problem:** Google and NVIDIA providers didn't have default model chains defined.

**Solution:**
- ✅ Added `buildGroqPlanChain()` with Llama and Mixtral models
- ✅ Added `buildGooglePlanChain()` with Gemini models  
- ✅ Added `buildNvidiaPlanChain()` with Llama models
- ✅ Updated `defaultPlanningChainFromSettings()` to use provider-specific chains
- ✅ All providers now have proper fallback model chains

## Files Modified

1. **chat/server.js**
   - Fixed Groq API implementation (lines ~3405-3450)
   - Added model to Mistral API response (line ~3348)
   - Added model to Google API response (line ~3402)
   - Added model to NVIDIA API response (line ~3479)
   - Added `buildGroqPlanChain()` function (lines ~3097-3104)
   - Added `buildGooglePlanChain()` function (lines ~3106-3113)
   - Added `buildNvidiaPlanChain()` function (lines ~3115-3120)
   - Updated `defaultPlanningChainFromSettings()` (lines ~2007-2033)

2. **scripts/entrypoint.sh**
   - Removed log file redirection (line 187)
   - Changed from: `node ... >/var/log/chat-service.log 2>&1 &`
   - Changed to: `node ... &`

## Testing Instructions

### Test 1: Verify Container Logs Work
```bash
# Start the container
docker-compose up -d

# Tail the logs - you should now see application output
docker logs -f shopify-ai-builder

# Look for logs like:
# [2024-01-11T...] Server started on http://0.0.0.0:4000
# [CONFIG] OpenRouter: { configured: true, ... }
# [CONFIG] Mistral: { configured: true, ... }
```

### Test 2: Verify Mistral Planning Works
1. Set your `MISTRAL_API_KEY` in environment variables
2. Go to Admin Panel → Plan models
3. Set "Mistral" as the primary planning provider
4. Save the configuration
5. Go to the builder and create a new project
6. Enter a planning request (e.g., "Create a WordPress plugin for contact forms")
7. Check that you receive a response
8. Check `docker logs` for Mistral-related logs with `[MISTRAL]` prefix

### Test 3: Verify Groq Planning Works
1. Set your `GROQ_API_KEY` in environment variables
2. Go to Admin Panel → Plan models
3. Set "Groq" as the primary planning provider
4. Save the configuration
5. Go to the builder and create a new project
6. Enter a planning request
7. Check that you receive a response
8. Check `docker logs` for Groq-related logs with `[GROQ]` prefix

### Test 4: Verify Provider Fallback
1. Configure multiple providers in the planning chain
2. Intentionally use an invalid API key for the first provider
3. Make a planning request
4. Verify that it automatically falls back to the next provider
5. Check logs to see the fallback chain in action

## Environment Variables Required

### For Mistral Planning
```env
MISTRAL_API_KEY=your_mistral_key_here
MISTRAL_API_URL=https://api.mistral.ai/v1/chat/completions  # Optional, uses default if not set
```

### For Groq Planning
```env
GROQ_API_KEY=your_groq_key_here
GROQ_API_URL=https://api.groq.com/openai/v1/chat/completions  # Optional, uses default if not set
```

### For Google Planning (if using)
```env
GOOGLE_API_KEY=your_google_key_here
```

### For NVIDIA Planning (if using)
```env
NVIDIA_API_KEY=your_nvidia_key_here
```

## Admin Panel Configuration

### Setting Up Planning Providers

1. **Access Admin Panel:**
   - Navigate to `/admin/login`
   - Log in with admin credentials

2. **Configure Planning Priority:**
   - Go to "Plan models" section
   - You'll see a list of planning models with priority
   - Click "Add planning model" to add providers
   - Drag to reorder (highest priority first)
   - Each row should specify:
     - Provider (openrouter, mistral, groq, google, nvidia)
     - Model (optional - uses defaults if not specified)

3. **Configure Rate Limits:**
   - Set tokens per minute/day limits per provider
   - Set requests per minute/day limits per provider
   - Monitor live usage in the same panel

## Default Models

### Groq
- Primary: `llama-3.3-70b-versatile`
- Fallback 1: `mixtral-8x7b-32768`
- Fallback 2: `llama-3.1-70b-versatile`

### Mistral
- Uses models configured in admin panel
- Default: `mistral-large-latest`

### Google
- Primary: `gemini-1.5-flash`
- Fallback 1: `gemini-1.5-pro`
- Fallback 2: `gemini-pro`

### NVIDIA
- Primary: `meta/llama-3.1-70b-instruct`
- Fallback: `meta/llama-3.1-8b-instruct`

## Logging Details

### Log Prefixes
All logs use consistent prefixes for easy filtering:
- `[MISTRAL]` - Mistral API operations
- `[GROQ]` - Groq API operations
- `[PLAN]` - Plan message handling
- `[CONFIG]` - Configuration at startup

### Viewing Specific Logs
```bash
# View only Mistral logs
docker logs shopify-ai-builder 2>&1 | grep "\[MISTRAL\]"

# View only Groq logs
docker logs shopify-ai-builder 2>&1 | grep "\[GROQ\]"

# View only planning logs
docker logs shopify-ai-builder 2>&1 | grep "\[PLAN\]"

# View configuration logs
docker logs shopify-ai-builder 2>&1 | grep "\[CONFIG\]"
```

## Verification Checklist

- [ ] Container logs are visible using `docker logs`
- [ ] Server startup logs show provider configuration
- [ ] Mistral planning requests return responses
- [ ] Groq planning requests return responses
- [ ] Provider fallback works when primary fails
- [ ] Admin panel shows all providers (openrouter, mistral, google, groq, nvidia)
- [ ] Rate limiting configuration works
- [ ] Usage statistics display correctly

## Known Limitations

1. **Google and NVIDIA APIs**: The current implementations use placeholder endpoints. These will need to be updated with the actual API endpoints and request formats if you plan to use them.

2. **Model Discovery**: Some providers may not support automatic model discovery. You may need to manually specify model names in the admin panel.

3. **API Key Validation**: API keys are not validated on configuration. Invalid keys will only be detected when making actual API calls.

## Troubleshooting

### Issue: Still No Logs Visible
**Solution:** Make sure to rebuild the container after pulling the changes:
```bash
docker-compose down
docker-compose build --no-cache
docker-compose up -d
```

### Issue: Planning Returns Error "API key not configured"
**Solution:** Ensure environment variables are properly set in your `.env` file or `docker-compose.yml`

### Issue: Planning Returns No Response
**Solution:** 
1. Check container logs for detailed error messages
2. Verify API key is valid
3. Check if provider has rate limits or is down
4. Try configuring a fallback provider

### Issue: Groq Returns Invalid Model Error
**Solution:** The default models should work, but if you get this error, check Groq's documentation for current model names and update the model chain in admin panel.

## Support

If you encounter issues:
1. Check the container logs first: `docker logs -f shopify-ai-builder`
2. Look for error messages with provider-specific prefixes
3. Verify your API keys are valid
4. Check the admin panel configuration
5. Try the fallback chain with multiple providers configured