Magentic provides decorators to create LLM-backed Python functions with structured outputs, streaming, and multi-backend configuration. It supports OpenAI, Anthropic, LiteLLM, and Mistral backends, with recent releases introducing streaming and OpenRouter support.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
Magentic is a Python library to integrate Large Language Models into Python code. It uses the @prompt and @chatprompt decorators to return structured outputs using pydantic models or built-in types, enabling function calls, streaming, and agent-like behavior. It includes features for observability and retries, and supports multiple backends for LLM providers.
How it works
The library exposes decorators (@prompt, @chatprompt, @prompt_chain) to define LLM prompts and chat flows. It can perform function calling, streaming outputs, and iterative prompts. It supports Configured backends (OpenAI, Anthropic, LiteLLM, Mistral) with environment/config options to switch providers. Examples show creating prompts that map to pydantic models and streaming results via StreamedStr. The project mentions FunctionCall objects and the ability to pass resulting calls to other decorators.
Getting started
Install via:
pip install magentic
or using uv
uv add magentic
Configure the OpenAI API key via OPENAI_API_KEY. For other backends, see configuration docs.
Usage highlights include:
- Using @prompt to define prompts that return typed outputs
- Using @chatprompt to define chat-style prompts with system and user messages
- Using @prompt_chain to resolve FunctionCall objects automatically
Recent releases
- v0.41.1 (2026-03-11): Handle empty stream chunks & remove require_parameters in OpenRouterChatModel by @piiq in https://github.com/jackmpcollins/magentic/pull/464
- v0.41.0 (2025-10-14): Add support for verbosity and max_completion_tokens request parameters to OpenAI chat model; Bump astral-sh/setu
- v0.40.0 (2025-06-22): Bump astral-sh/setup-uv; Add OpenRouter chat model
- v0.39.3 (2025-04-07): Fix: function call parsing positional args ignoring arg defaults
- v0.39.2 (2025-03-02): Add tests for Gemini via openai package
Traction
Stars: 2415, Forks: 127, Open issues: 49
Behind the repo
Not included in the provided facts.
Caveats
License: MIT. Created: 2023-06-18. Last push: 2026-03-11. Language: Python.