LLM Migrator

Version 1- OpenAI → Mistral Migration Tool

Automatically analyze your codebase and generate migration guides to switch from OpenAI to Mistral AI.

Scan your project, identify OpenAI API usage, and get side-by-side code comparisons with cost savings analysis.

---

Quick Start

# Scan your codebase
python migrate_to_mistral.py scan ./src

# Save migration guide to file
python migrate_to_mistral.py scan ./src --output migration_guide.txt

# Try the web UI
python3 app.py
# Visit http://localhost:5000

---

Features

• 🔍 Automatic Detection - Scans Python codebases for OpenAI API patterns
• 💰 Real-Time Pricing - Fetches current model pricing from OpenRouter API
• 📊 Accurate Cost Analysis - Calculates real savings based on detected models and usage
• 📝 Migration Examples - Generates side-by-side code comparisons
• 🌐 Web Interface - Modern UI with GitHub repository scanning
• 🎯 Effort Estimates - Shows migration complexity for each pattern
• 📁 File Mapping - Lists exact files and line numbers to update
• 🤖 Model Detection - Identifies which GPT models you're using (GPT-4, GPT-3.5, etc.)

---

Example Output

🔍 Scanning ./examples for OpenAI API calls...

================================================================================
🚀 MIGRATION GUIDE: OpenAI → Mistral AI
================================================================================

📊 SUMMARY
Found 5 OpenAI API call(s) across 1 file(s)

Pattern breakdown:
  • chat: 2 call(s)
  • embedding: 2 call(s)
  • streaming: 1 call(s)

💰 ESTIMATED SAVINGS
Assuming moderate usage (10M tokens/month):
  OpenAI (gpt-4): ~$300/month
  Mistral (mistral-large): ~$100/month
  📉 Savings: ~$200/month (67% reduction)

📝 MIGRATION EXAMPLES

## CHAT MIGRATION
Effort: trivial

BEFORE (OpenAI):
----------------------------------------
from openai import OpenAI
client = OpenAI(api_key="your-key")

response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Hello!"}]
)

AFTER (Mistral):
----------------------------------------
from mistralai.client import MistralClient
client = MistralClient(api_key="your-key")

response = client.chat(
    model="mistral-large-latest",
    messages=[{"role": "user", "content": "Hello!"}]
)

---

Supported Patterns

The tool identifies these OpenAI API patterns:

Pattern	Difficulty	Savings
Chat Completions	Trivial	70%
Streaming	Trivial	70%
Function Calling	Trivial	70%
Embeddings	Easy	80%

---

Web Interface

Launch the web UI for an interactive experience:

# Install Flask (only dependency for web UI)
pip install flask

# Start the server
python3 app.py

Features:

• Local Directory Scanning - Analyze projects on your machine
• GitHub Repository Scanning - Clone and analyze any public repo
• Real-Time Pricing - Live cost calculations from OpenRouter API
• Visual Cost Comparison - Interactive charts and statistics
• Model Detection - Shows which GPT models are in use
• Downloadable Reports - Export migration guides

Real-Time Pricing Integration

The tool now fetches live pricing data from OpenRouter to calculate accurate cost savings:

# Automatically detects your model usage
# Calculates real costs based on:
# - Detected GPT models (GPT-4, GPT-3.5, etc.)
# - Current market pricing
# - Estimated token volumes

Example Output:
  OpenAI (GPT-4): $450/month
  Mistral (Large): $40/month
  Savings: $410/month (91.1%)

See PRICING_INTEGRATION.md for full documentation.

---

Installation

# Clone the repository
git clone https://github.com/yourusername/openai-to-mistral
cd openai-to-mistral

# No dependencies for CLI! Uses only Python stdlib
python migrate_to_mistral.py scan ./your-project

# For web UI, install Flask
pip install flask

---

Migration Example: RAG Application

Before (OpenAI):

from openai import OpenAI
import pinecone

client = OpenAI()

# Generate embedding
embedding = client.embeddings.create(
    model="text-embedding-ada-002",
    input=query
).data[0].embedding

# Chat with context
response = client.chat.completions.create(
    model="gpt-4",
    messages=[
        {"role": "system", "content": "Answer based on context"},
        {"role": "user", "content": f"Context: {results}\n\nQuestion: {query}"}
    ]
)

After (Mistral):

from mistralai.client import MistralClient
import pinecone

client = MistralClient()

# Generate embedding
embedding = client.embeddings(
    model="mistral-embed",
    input=[query]  # Note: wrap in list
).data[0].embedding

# Chat with context
response = client.chat(
    model="mistral-large-latest",
    messages=[
        {"role": "system", "content": "Answer based on context"},
        {"role": "user", "content": f"Context: {results}\n\nQuestion: {query}"}
    ]
)

Changes:

1. Import statement (1 line)
2. Embedding method (1 line)
3. Chat method (1 line)
4. Model names (2 places)

Time: 5 minutes Savings: ~70% on inference costs Risk: Low (API is 95% compatible)

---

Model Mapping

OpenAI Model	Mistral Equivalent	Use Case
gpt-4 / gpt-4-turbo	mistral-large-latest	Complex reasoning, code
gpt-3.5-turbo	mistral-small-latest	Simple tasks, high volume
text-embedding-ada-002	mistral-embed	Embeddings for RAG

---

Cost Comparison (Per 1M Tokens)

Chat Completions

Model	OpenAI Price	Mistral Price	Savings
High-end	$30 (GPT-4)	$10 (Large)	67%
Mid-tier	$10 (GPT-4 Turbo)	$4 (Large)	60%
Volume	$1.50 (GPT-3.5)	$0.65 (Small)	57%

Embeddings

Model	OpenAI Price	Mistral Price	Savings
Standard	$0.13 (Ada-002)	$0.02 (mistral-embed)	85%

Real-World Example

Scenario: AI support chatbot processing 10M tokens/month

• OpenAI (gpt-4): $300/month
• Mistral (large): $100/month
• Annual savings: $2,400

For a startup with 5 AI features, that's $12k/year in savings.

---

Architecture

migrate_to_mistral.py
├── MigrationAnalyzer      # Scans code for OpenAI patterns
│   ├── Pattern matching (regex)
│   ├── AST analysis for context
│   └── Confidence scoring
│
└── MigrationGuideGenerator # Creates actionable guides
    ├── Code diffs (before/after)
    ├── Cost calculations
    └── Effort estimates

app.py                     # Flask web application
├── GitHub repository cloning
├── Temporary file management
└── REST API endpoints

---

Limitations

• Python only - JavaScript/TypeScript support planned for v2
• Pattern-based detection - May miss dynamic or obfuscated usage
• No auto-rewriting - Generates guides for manual migration (safer)
• Requires testing - Always test migrated code thoroughly

---

Roadmap

v1.0 (Current)

• ✅ Python codebase scanning
• ✅ Pattern detection for chat, streaming, embeddings
• ✅ Side-by-side code examples
• ✅ Cost calculations
• ✅ Web UI with GitHub scanning

v2.0 (Planned)

• JavaScript/TypeScript support
• Auto-generate PR with migrations
• Integration tests generation
• Performance comparison reports

v3.0 (Future)

• CI/CD integration
• Gradual rollout strategies
• A/B testing framework
• Real-time cost monitoring

---

Contributing

Contributions are welcome! Here's how you can help:

• Report bugs - Open an issue with reproduction steps
• Suggest patterns - Help us detect more OpenAI API patterns
• Add features - Submit PRs for new functionality
• Improve docs - Better examples, clearer explanations

---

License

MIT License - feel free to use, modify, and share.

---

Try it now: python migrate_to_mistral.py scan ./your-code