Mixtcha allows you to effortlessly blend multiple LLMs (GPT-4, Claude, etc.) into a single API call. Run models in parallel, aggregate their responses, and get high-quality results.

1. Get your API key from mixtcha.com and add some credits to your Mixtcha wallet.

2. Save the API key in .env:

   MIXTCHA_API_KEY=sk-mix-1234-your-key
   

3. Install dependencies:

   pip install openai python-dotenv pyyaml

4. Start mixing LLMs:

   import os
   import dotenv
   import openai
   import yaml
   
   dotenv.load_dotenv()
   client = openai.OpenAI(
       api_key=os.getenv("MIXTCHA_API_KEY"),
       base_url="https://api.mixtcha.com"
   )
   
   # Define a parallel + aggregator configuration
   config = {
       "layers": [
           {
               "type": "parallel",
               "models": ["openai/gpt-4o", "anthropic/claude-3.5-sonnet"]
           },
           {
               "type": "aggregator", 
               "model": "anthropic/claude-3.5-sonnet",
               "prompt": "Synthesize the previous responses into a single, high-quality answer."
           }
       ],
       "messageMode": "inline",
       "delimiter": ["<option>", "</option>"]
   }
   
   response = client.chat.completions.create(
       model=yaml.dump(config),
       messages=[
           {"role": "user", "content": "Why are mixtures of LLMs powerful?"}
       ]
   )
   
   print(response.choices[0].message.content)

• Parallel Processing: Run multiple LLMs simtaneously
• Smart Aggregation: Synthesize multiple responses into one
• Simple API: Uses familiar OpenAI-style interface, use mixtcha with any application or client that allows you to override the base_url and model
• Flexible Configuration: Define custom LLM mixtures in YAML/JSON
• Transparent: Inspect intermediate responses from each model, including the cost

• Check out our Quickstart Notebook for a detailed walkthrough
• Browse available models and pricing at mixtcha.com/models_list.yaml
• See our TypeScript type definitions for type-safe configurations
• Save your mixtcha configurations as YAML/JSON files and share them with the community.