magentic

Why we included this project

magentic lets you express LLM calls as ordinary Python functions. You add a @prompt or @chatprompt decorator, specify a return type like a pydantic model or a built-in list, and the library handles the prompt template, the model request, and parsing the structured output. The decorated functions stay callable like any other function, so you can chain them, pass them to other prompts, or let @prompt_chain resolve function calls into multi-step workflows. Streaming, async support, and OpenTelemetry tracing fit it into production pipelines where you want partial results and visibility into what the model did. If you already use pydantic and value type checking and IDE hints, the design will feel natural, and switching between OpenAI, Anthropic, and Ollama does not require rewriting your functions.