TypeScript's pick-and-mix AI toolkit

What it does

Axflow is a collection of TypeScript modules for building natural-language applications. You can adopt pieces incrementally—an SDK for model calls, a data-to-LLM connector, and an evaluation framework—rather than swallowing a monolithic platform whole.

The interesting bit

The “code-first” pitch is the real angle: they want to deconstruct LLM complexity into composable parts you actually control, not hide it behind a GUI. The zero-dependency models package with React hooks and streaming utilities suggests they understand the frontend pain points.

Key highlights

• @axflow/models — zero-dependency SDK with React hooks and streaming utilities
• axgen — framework for connecting your data to LLMs
• axeval — framework for evaluating LLM output quality
• Three more modules in progress: extract (document processing), serve (LLM hosting with middleware), finetune
• MIT licensed, 1,120 stars

Caveats

• Several promised modules (extract, serve, finetune) are listed as “working on” with no clear timeline
• The README leans aspirational; concrete benchmarks or comparison to alternatives aren’t present
• The demo UI and video showcase axgen specifically, not the full framework

Verdict

Worth a look if you’re building TypeScript AI apps and want modular pieces you can swap, not a vendor lock-in. Skip it if you need a fully-baked, batteries-included platform today—the framework is still filling in its gaps.