Actionengine

[](https://github.com/JetBrains/.github/blob/main/profile/README.md)

Action Engine

Action Engine is a cross-language toolkit for building interactive applications that need streaming, multimodal, interactive capabilities. Think of:

• GenAI prototypes and fully served applications that need different media modalities and interactivity—image generation and editing, speech to text, etc.
• LLM agents that can interact with users and tools
• Native applications that need to interact with LLMs or just combine API calls asynchronously in a flexible way

The core library is written in C++, and there are bindings for Python with almost complete feature coverage. A TypeScript library is also available for building UIs that interact with the core library.

Quickstart in Python

If there is a wheel already built for your platform, you can install Action Engine from PyPI (note that the package is still under development, so the API is not stable yet):

pip install action-engine
python -c 'import actionengine; print(actionengine.to_chunk("Hello, world!"))'

Otherwise, you can install Action Engine directly from the GitHub repository. First, make sure you have a recent C++20-compatible `clang` installed, then run:

pip install git+https://github.com/JetBrains/actionengine -v
python -c 'import actionengine; print(actionengine.to_chunk("Hello, world!"))'

You can then explore the examples in the `examples` folder, such as [007-python-generative-media](examples/007-python-generative-media), which showcases how to build generative media applications with LLMs and text-to-image models.

Action Engine requires Python 3.11 or higher.

Features

• Flexible unopinionated architecture based on two core concepts:
  • async nodes, which are not more and not less than asynchronous data channels, with a bring-your-own storage system in mind,
  • actions, which are computations