Settings

User → Settings, or Ctrl+,. The same shortcut closes it again.

It opens as a floating window with six tabs — General, Models, MCP Servers, Instructions, Keybindings and Privacy — and it is not a modal. The IDE behind it stays live: you can type in the editor, read the chat, watch the agent work, all with Settings sitting open on top. Drag it out of the way rather than closing it.

General, Keybindings and Privacy apply the moment you change them — no Save button. Models, MCP servers and Instructions are edited in forms with Save and Cancel, and nothing you type there is stored until you press Save. Nothing in the window needs a restart.

Where your settings live — the part that surprises people#

The six tabs look like one thing. They are not: settings in this window are stored in four different places, with four different ideas of who they belong to. Which one a setting uses decides whether it follows you to your next project, to your next machine, or nowhere at all.

Per project. Everything in General, and all of your Keybindings. These live beside that project's index, so they're remembered per project and only per project. Set your editor to 15pt in one project and the next one still opens at 13. Rebind Ctrl+J here and it's still Ctrl+J there.

Per machine, all projects. Your custom models, your MCP servers, your global instruction, and your user-scope slash commands. These are files under ~/.semantix/USER/, so they apply in every project you open on this computer — and they don't travel to another one.

Per project, as files you can edit. Project instructions and workspace slash commands — markdown files under ~/.semantix/WORKSPACE/<project-id>/, readable and editable outside the app.

→ Instructions and commands

Per account, everywhere. Only Privacy. Those four switches are stored against your account, so they follow you to any machine you sign in on.

The Instructions and MCP tabs tell you their scope on screen — a Workspace / User toggle, an Auto · all projects badge, the storage path printed under the heading. General, Keybindings and Privacy don't say anything, which is why this is worth reading once. When a setting mysteriously "didn't stick", it almost always stuck — in a scope you weren't looking at.

The six tabs#

General — font sizes for the interface, the editor and the chat; the editor's syntax theme; a switch that turns the 3D view off entirely; and the Re-index project button, which is the fix for the IDE having missed changes you made while it was closed.

Models — point the agent at any OpenAI-compatible or Anthropic-compatible endpoint of your own. Base URL, model id, key, and optional pricing so cost readouts work.

MCP servers — connect external tool servers. Anything they expose becomes available to the model on every lane, named mcp__<server>__<tool>.

Instructions and commands — two features sharing one tab. Standing instructions that ride along with your messages, and reusable /slash prompt templates.

Keybindings — see the app's shortcuts and rebind about half of them; the rest are listed but fixed. Also worth reading for the shortcuts that work but aren't in there at all.

Privacy — what gets stored, what gets backed up, and what you can turn off.

What isn't here#

Theme and colours. There's no Theme tab, on purpose — editing the look of the IDE is a whole surface of its own, reached from User → Theme Editor. The interface typeface and the code font both live there too, not in General.

→ The theme editor

Choosing a model for a conversation. The Models tab is where you define a model; the picker in the chat composer is where you choose one, per conversation.

→ Choosing the model

Your Claude sign-in. Nothing in this window touches it.

Small things about the window#

It opens centred at a fixed size every time — it doesn't remember where you left it or how big you made it. It also always opens on General, whichever tab you were last using.

You can drag it by its title bar, resize it from any edge or corner, minimise it to a bar in the bottom-left, or maximise it to fill the work area. With the window focused, Ctrl+← / Ctrl+→ step it left, centre and right, and Ctrl+↑ / Ctrl+↓ dock it to the top or bottom edge. Maximising and the arrow keys keep it inside the work area, and it can never be dragged up under the menu bar — but dragging it down will leave most of it hanging over the status bar.

A few actions close it for you, so the thing you just asked for isn't hidden behind it: switching to runs mode, signing in from the profile menu, and — in the normal chat shell only — opening a conversation from history or jumping to a tool call from a node card.