Adapters

Adapters handle the translation between Signatures and LLM-specific message formats.

Overview

The ChatAdapter is responsible for:

1. Converting signatures into system prompts
2. Formatting inputs into user messages
3. Parsing LLM completions into structured outputs
4. Converting Pydantic models to tool schemas

Usage

from udspy import ChatAdapter

adapter = ChatAdapter()

Adapters are typically used internally by modules, but can be used directly:

# Format instructions
instructions = adapter.format_instructions(signature)

# Format inputs
formatted = adapter.format_inputs(signature, {"question": "What is AI?"})

# Parse outputs
outputs = adapter.parse_outputs(signature, completion_text)

Custom Adapters

You can create custom adapters by subclassing ChatAdapter:

class CustomAdapter(ChatAdapter):
    def format_instructions(self, signature):
        # Custom instruction formatting
        return super().format_instructions(signature) + "\nBe creative!"

See API: Adapters for detailed documentation.