Reasoning Closed for llama.cpp Autor: Firedreams Studio

The llama.cpp web UI auto-expands a model's reasoning ("thinking") block while a reply streams in. For roleplay, storytelling, puzzles, or anything where you want to read the answer cold, that block spoils the result before you get to it.

Keep Reasoning Closed keeps that block collapsed by default, on every chat and every device, without changing anything on your server.

What it does

• Keeps the reasoning block closed while the model generates, so the answer is never spoiled.
• Leaves you in full control: click the "Reasoning" header any time to open it, and it stays open until you close it.
• Only closes blocks that opened on their own. A block you open by hand is never touched.

Works with your own server

llama.cpp is self-hosted, so you add your server's address on the extension's options page (for example http://192.168.1.10:8081). localhost and 127.0.0.1 work out of the box. The extension only ever runs on the exact addresses you add.

Private by design

No data is collected and nothing leaves your browser. The extension only requests access to the servers you explicitly add, and you can remove any of them at any time.