Ask HN: What are good AI UIs now?

[ramstar3000]

With frameworks like Streamlit, it takes five lines of Python to wrap an LLM in a chat box.

Alternatively, we've seen a surge in TUI tools (Claude Code, Codex, etc.). But living in a terminal doesn't feel like the final destination. Now, we are seeing wrapper GUIs like T3 Code to give those terminal tools structure.

What products or projects (if any) are rethinking human-AI interaction?

[paul_knoxops]

I believe the IM format is still the most acceptable and easiest to use for current users. It allows you to see the AI's working process with complete transparency, especially when the AI is working on long tasks.

Of course, some underlying implementations need to be hidden; users don't need to know everything, it depends on your product's design.

[__patchbit__]

The vertical bread board wiring on the experimental MIT AI Lab CADR LISP Machine suggests token streams should have trace patterns and then the hot path to focal result for prompt should look like a vertical with a turn when the AI model settles.

[Vignesh_Reddy]

Depends on the use case. If you're building production AI features, the UI is the least of your problems — cost attribution and hallucination detection across your LLM calls matter more than the chat interface.

[sometimelurker]

llama.cpp + llama-serve, I use it. if you like terminal, pi