Awesome AI Agents › AI Agent Development

jackmpcollins/magentic

⭐ 2425 Python repository created 2023-06-18

Magentic is a Python library designed to seamlessly integrate Large Language Models (LLMs) into Python code by allowing developers to create functions that return structured outputs from LLMs. It provides decorators such as `@prompt` and `@chatprompt` to define LLM prompts as Python functions, enabling the combination of LLM queries and traditional Python code to build complex agentic systems. The library supports structured outputs using Pydantic models and built-in Python types, ensuring that the data returned from LLMs adheres to defined schemas. It also supports streaming of outputs and function calls, allowing real-time processing of generated content. Magentic includes features like LLM-assisted retries to improve adherence to complex output schemas, observability through OpenTelemetry with native Pydantic Logfire integration, and type annotations for better compatibility with linters and IDEs. It supports multiple LLM providers including OpenAI, Anthropic, and Ollama, and offers additional capabilities such as chat prompting, parallel function calling, vision integration, formatting, and asyncio support. The `@prompt` decorator allows defining a prompt template as a Python function, which when called, sends the prompt to an LLM and returns the generated output, respecting the function's return type annotation. The `@chatprompt` decorator extends this by allowing chat message templates for few-shot prompting and system messages. Magentic also supports LLM-driven function calls where the LLM can decide to call specific functions, returning a `FunctionCall` object that can be executed with the provided arguments. For more complex workflows, the `@prompt_chain` decorator enables chaining multiple function calls resolved automatically by the LLM until a final answer is produced. The library also supports streaming outputs, allowing processing of text as it is generated, and concurrent streaming for multiple outputs. Magentic is installable via pip and configurable through environment variables for API keys and other settings. Overall, Magentic empowers developers to harness the power of LLMs within Python applications efficiently, providing structured, observable, and extensible integration for advanced AI-driven functionalities.

https://github.com/jackmpcollins/magentic

agentagenticaianthropicapi-integrationasynciochat-promptingchatbotchatgptchatprompt-decoratorfunction-callingfunction-callsgptlarge-language-modelsllmmagentamagenticmagneticobservabilityollamaopenaiopenai-apiopentelemetrypromptprompt-chainingprompt-decoratorpydanticpythonstreamingstructured-outputtype-annotations