llm-ui

Why we included this project

If you're building a React chat interface or streaming assistant, you've probably spent time wrestling with markdown renderers and choppy token streams. llm-ui handles that in-between layer for you: it cleans up broken markdown from raw model output, throttles stream pauses so text appears smoothly, renders at native frame rate, and lets you drop in your own components alongside Shiki-highlighted code blocks. Because it works on the model's output string rather than tying into a specific vendor SDK, it'll work with OpenAI, Claude, Ollama, Mistral, or Hugging Face, and it's headless, so you keep your own styling. For frontend engineers wiring an LLM into a React product, that means skipping a lot of hand-rolled parsing, buffering, and per-token animation logic.