Axflow is a collection of modules for building robust natural language powered applications. These modules can be adopted incrementally, thus providing a modular and scalable solution. Used together, they form an end-to-end framework for developing AI applications.

• @axflow/models — A zero-dependency, modular SDK for building robust natural language applications. Includes React hooks and streaming utilities that make building AI applications a breeze.
• axgen — A framework for connecting your data to large language models
• axeval — A framework for evaluating LLM output quality

In addition to the above modules, we're working on the following modules:

• extract: A library for efficient data processing, particularly loading, transforming, and chunking documents from arbitrary sources. Most useful for applications that need to load and preprocess data for vector search.
• serve: A serving framework to run any LLM model (OSS or otherwise). It will also provide middleware options for user throttling, analytics, and logging
• finetune: A library focused on fine-tuning models

Axflow aspires to deconstruct the complex paradigms of working with LLMs into manageable and intuitive components. Our library takes a code-first approach, emphasizing the importance of flexibility and control for developers. As a foundational framework, Axflow empowers developers to build higher-level TypeScript AI features and products seamlessly.

Here is an example open source UI showcasing what our first module, axgen, can do, with a short video walkthrough.

MIT