UI & Chat Interfaces AI Tools
User interfaces, chat UIs, and frontend tools for AI applications
A model endpoint is not a product. Once a local runtime or hosted API answers on a port, what remains is conversation state, streaming, history, tool call rendering, and some notion of who is logged in. These projects supply that front door. The field splits into three camps in 2026. Self-hosted chat servers such as Open WebUI, LibreChat, Lobe Chat, and Khoj ship a finished multi-user application with retrieval, provider routing, and auth already wired. Desktop clients like Chatbox and Cherry Studio drop the server and keep keys and transcripts on one machine. Developer libraries hand over primitives instead: assistant-ui and CopilotKit for embedding chat in an existing codebase, Gradio and Streamlit for quick Python demo surfaces. The tradeoff is ownership of that surface. Finished apps run the same afternoon but impose their data model, their auth, and their upgrade cadence, while primitives require building the shell and in return let chat live inside a product rather than beside it. A sensible starting order: Open WebUI for a first self-hosted deployment over local models, Gradio for a demo around one Python function, assistant-ui once chat must sit inside an existing React codebase. The trap here is license drift. Several widely deployed chat frontends relicensed after gaining traction, layering branding-retention or commercial-use conditions onto code that began as MIT or Apache, usually triggered above a user count or when the logo is stripped. The LICENSE file at the commit being deployed is the authority, not the badge in the README.