LangStream is a Python library designed for building robust LLM applications with a focus on composability and a minimal core. It offers a lighter alternative to frameworks like LangChain, enabling developers to create complex LLM workflows by composing simple, type-hinted "Streams."

How It Works

LangStream's core abstraction is the Stream, which represents a single operation, such as an LLM call, data parsing, or prompt formatting. These streams are built using Python's AsyncGenerator and can be composed sequentially using methods like and_then, map, collect, join, and gather. This functional programming approach, combined with strict type signatures for stream inputs and outputs, promotes code clarity, reduces bugs, and simplifies debugging by allowing access to intermediate results.

Quick Start & Requirements

• Install: pip install langstream openai
• Prerequisites: Python 3.7+, OpenAI API key.
• Documentation: https://langstream.org/docs/
• Example: https://github.com/rogeriochaves/langstream/blob/main/examples/chatbot.py

Highlighted Details

• True Composability: Small set of functional composition operators (and_then, map, collect, join, gather) for building complex pipelines.
• Type Safety: Explicit input/output type signatures for streams prevent integration errors and enable editor-level validation.
• External Prompts: Encourages keeping prompts outside of core logic for easier customization and LLM switching.
• Bring Your Own Integration: Designed for easy addition of custom integrations with minimal abstraction layers.

Maintenance & Community

• Active development, seeking contributions for integrations and documentation.
• Discord community available for support and updates.
• Roadmap includes memory mechanisms, RxPY integration, and LangChain compatibility.

Licensing & Compatibility

• License: MIT.
• Compatible with commercial and closed-source applications.

Limitations & Caveats

The project is relatively new, with ongoing efforts to expand integrations and documentation. Some planned features, like a default memory mechanism, are still under development.