--- title: "Settings" description: "Use Settings when you want to make the terminal fit your hands: choose a theme, connect an AI provider, tune suggestions, add skills, or change shortcuts." canonical: "https://con.nowledge.co/docs/settings/" source: "https://github.com/nowledge-co/con-terminal/blob/main/docs/settings.md" keywords: "terminal emulator, AI terminal, terminal-first AI agent, terminal agent, AI agent terminal, agentic terminal, agent-native workflows, open-source terminal, open-source AI terminal, Warp alternative, Warp terminal alternative, terminal for coding agents, CLI agent workflows, SSH AI terminal, tmux AI terminal, GPU terminal, Rust terminal, native terminal app, developer tools, AI providers, OpenAI terminal, Anthropic terminal, DeepSeek terminal, xAI terminal, GitHub Copilot terminal" --- # Settings con should feel good before you change anything. Use Settings when you want to make the terminal fit your hands: choose a theme, connect an AI provider, tune suggestions, add skills, or change shortcuts. Open Settings from the app menu on macOS, the gear button on Windows and Linux, or the Command Palette. Appearance changes apply as you tune them, so you can leave Settings open while checking opacity, blur, fonts, and background images. ## General General contains app-level behavior: - update channel and version - manual update check when the current channel supports it - saved terminal text privacy - skill source folders If you prefer con not to save terminal text between launches, turn off **Restore Terminal Text** in General. Layout profiles never include terminal text. To wipe terminal text already saved on disk, run **Clear Restored Terminal History** from Command Palette. ## Command Palette Press ⌘ ⇧ P on macOS, or ⌃ ⇧ P on Windows and Linux. The Command Palette is the fastest way to find actions you do not use every minute. Search for Settings, workspace profiles, pane actions, surface actions, updates, privacy actions, and other commands. When an action has a shortcut, con shows it beside the action. ## Appearance Appearance controls the parts of con you look at all day:

Con Appearance settings with theme and transparency controls

- terminal theme - terminal and UI fonts - terminal opacity and blur - background image - tab position - pane title bars in split layouts Start with readability. Pick a theme with clear contrast, then adjust opacity or blur only if the terminal remains easy to scan. Pane title bars appear only when a tab has multiple panes. Keep them on if you want direct close/fullscreen controls and drag-to-rearrange. Turn them off if you prefer a sparse terminal surface and use shortcuts or the terminal context menu for pane actions. con can import Ghostty themes. Copy a theme, choose **Load from Clipboard**, preview it, then save it when it feels right. ## AI providers The Providers section stores the connection details for the model hosts you use. The AI section chooses the active provider and model for the agent panel, Command Palette AI actions, and AI fallback suggestions. con supports Anthropic, OpenAI, ChatGPT, GitHub Copilot, OpenAI-compatible hosts, MiniMax, Moonshot, Z.AI, DeepSeek, Groq, Gemini, Ollama, OpenRouter, Mistral, Together, Cohere, Perplexity, and xAI. Use the strongest model for agent work when the answer matters. Use a faster or cheaper model for suggestions if you want inline help without slowing down the terminal. ChatGPT and GitHub Copilot can use OAuth, so you can leave the API key empty for those sign-in flows. OpenAI-compatible hosts can fetch models from `/models` when the host supports it. If the host has no models endpoint, type the model ID manually and save it. The provider picker in the agent panel shows configured providers. If a provider is missing there, configure it first in Settings. ## Tool approval The agent should act in view. Leave tool approval on when you are in an unfamiliar repository, a production shell, or any session where mistakes are expensive. Auto-Approve Tools lets the agent run tools without asking for each action. Use it only in workspaces where you trust the task, the model, and the recovery path. ## Command suggestions Suggestions are terminal help, not a second prompt. con checks local command history first. If history has no strong match, AI Command Suggestions can ask the configured suggestion provider for a fallback. You can turn this off, or route suggestions to a different provider and model from the main agent. Treat ghost text as a proposal. Accept it when it is what you meant, ignore it when it is not. ## Skills Skills are slash commands backed by a `SKILL.md` file. They are useful for work you repeat: release checks, project playbooks, debugging routines, writing rules, or team-specific workflows. Type `/` in the input bar or agent panel to browse available skills. Project skills live with the current workspace: - `skills/` - `.agents/skills/` - `.con/skills/` Global skills follow you across projects: - `~/.config/con/skills` - `~/.agents/skills` On Windows, the config skills folder is `~/.config/con-terminal/skills`. Keep project skills for shared project habits. Keep global skills for personal habits. If names collide, the project skill wins so a repository can define its own local meaning. For the workflow loop, see [Skills and workflows](https://con.nowledge.co/docs/skills-and-workflows/). ## Shortcuts The Keys section shows editable shortcuts for app, pane, and surface actions. It also includes the optional global Summon / Hide Con shortcut. That shortcut is off by default because global hotkeys can conflict with launchers and window managers. On macOS, Keys also includes Quick Terminal. It is off by default. When enabled, it opens a dedicated floating Con window from anywhere, separate from the main window. The default shortcut is ⌘ Backslash, and you can record a different one if it conflicts with your setup. For the full behavior, including hide/show state and how it differs from Summon / Hide Con, see [Quick Terminal](https://con.nowledge.co/docs/quick-terminal/). Change shortcuts when the default conflicts with muscle memory. Leave them alone when the built-in flow already works. A good setup should reduce decisions, not create a private language you have to remember.