If you're happy running local models, llama.cpp's built-in web-server's interface can do this.