TypeChat

Why we included this project

Getting an LLM to return structured data an app can actually use usually means writing and re-writing prompt templates until the model behaves. TypeChat flips that: you define the intents you want to support as TypeScript types, and the library builds the prompt from those types, validates that the model's reply conforms to the schema, and calls the model again to repair output that does not fit. So the developer's real work is deciding the shape of the data they want back, not tuning fragile prompts. The same approach works across TypeScript, Python, and C#/.NET. If you are building a natural language interface that needs well-typed responses, whether it is a sentiment classifier or a shopping assistant, this gives you a concrete, testable way to get there.