perfect ！

[orange0115]

awesome~
Recently, we also have some ideas, such as supporting conversational editing of spec proposals in the UI.

[Gaubee]

Yes, I’ve had similar ideas and plans.

Introduce interactive capabilities to revise the spec: e.g., select a chunk of text and let the AI rewrite it, or do a full-document polish, or auto-translate, etc.

I even fantasised about assigning AIs directly inside the UI and watching their status and progress in real time—like a MOBA game. GUIs excel at this kind of fine-grained control.

But I have to weigh whether bolting these AI features on rigidly will make the whole project uncontrollable. I need to think about how generalisable the value really is.

---

My current line of thought is “Web → MCP or Agent or Skills”(like npx openspecui mcp):
Let developers drive the web with their own agents. This way they can keep using whatever agent they like, or simply use the default agent we provide for OpenSpecUI.
OpenSpecUI itself stays focused on responsive, visual presentation of the spec.

Architecturally this is cleaner—no tight coupling inside OpenSpecUI—so I started a separate project, cenai, where I’m drafting the spec-prompt.

---

In the short term, however, just sprinkling a chat-AI into OpenSpecUI isn’t worth much. For developers, letting their own agent tweak the spec via a prompt already covers 80 % of cases.

OpenSpecUI’s biggest edge today is live spec rendering. Dropping in a chat-AI might nudge the experience from 80 % to 85 %.

So either don’t do it, or do it right. I’ll wait until cenai moves further along.

---

If you see it differently, I’d love to hear your take.

[waterbang]

I think this feature is too bloated to add. @orange0115

[1sh-io]

Perfect! I had the same idea as well—you just got there first.