All pages
Docs · Start here
Getting started
Availability: macOS. Linux and Windows run the GTK app, which has a smaller surface — see Platforms.
The first time you launch TRMNLY, a short tour walks you through the surfaces. You can replay it any time from Help → Show First-Run Tour.
Sign in
TRMNLY is in private testing. Release builds open only when signed in to a trmnly.dev account that has been approved, so the first launch shows a Sign in to use TRMNLY window before any terminal:
- Continue with GitHub copies a one-time code and opens GitHub's device page.
- Sign in with trmnly.dev opens a pairing page where you can create an account or log in, then confirm the code shown in the window.
A new account shows Waiting for approval until it's approved. The window checks every 30 seconds and unlocks on its own. Signing in here doesn't turn on remote control.
Once approved, TRMNLY renews your access in the background. It keeps working offline for 7 days since it last reached trmnly.dev. After that it asks you to reconnect. If your access is removed, TRMNLY asks you to save your work and quit. Unpairing in Settings → AI Agent → Remote control signs you out and quits.
The trmnly-code CLI uses the app's sign-in: if the app isn't signed in,
the CLI says so and exits.
Builds you compile yourself (Debug, ReleaseLocal) don't ask you to sign in.
Pick a model provider
The agent works with Anthropic Claude, OpenAI (and any OpenAI-compatible
/v1 endpoint — LM Studio, vLLM, etc.), Google Gemini, OpenRouter, and
local Ollama. Add a key from Settings → API Keys (Anthropic, OpenAI,
Gemini) or from the agent sidebar's provider sheet, which also holds the
OpenRouter key and local Ollama / LM Studio endpoints. Keys live in the
macOS Keychain — TRMNLY never writes them to disk.
In production builds, the Cred wallet (Settings → Account) sits in front of the agent regardless of which keys you've added; sign in and top up a balance before the agent runs. Dev builds without Cred backend config bypass the wallet entirely. See Cred wallet.
If you don't add a key, the rest of the terminal works as a normal Ghostty — splits, tabs, themes, fonts, and config all behave the same. Only the Agent tab is gated on a provider.
The three sidebars
TRMNLY keeps the terminal itself uncluttered and pushes feature surfaces into sidebars you can toggle:
| Sidebar | Toggle | Contents |
|---|---|---|
| Left | ⌘\ |
Projects rail + tab list, worktrees, recent folders, find-in-project |
| Right | ⌘⇧\ |
Agent, Prompts, Queue, Files, Web — plus a More menu for Tasks, Notes, Jobs, Conn, Keys, Docker, Devices, Flows, Local LLM, Activity, Plugins, Help |
| Quick Terminal | bind toggle_quick_terminal |
A floating HUD terminal that overlays any space |
You'll spend most of your time in the right sidebar. The segmented control at the top picks which tab is showing; the choice survives relaunch.
The command palette
⌘⇧P opens a palette that searches every menu action, open tabs and
panes, settings you can change inline (theme, font size), and agent
actions. If you can't remember a keybind, type a verb (split, theme,
tour, help) and it finds the command. ⌘K opens the lighter Quick
Switcher for jumping between tabs, projects, files, and recent
commands.
Two more chords worth learning early: ⌃⌘S grabs a screenshot — from any
app — and attaches it to the agent chat. ⌃⌘A (or the paperclip button
above the terminal) opens a file picker and types what you pick at the
cursor, so handing a file to a shell or a CLI agent never means a detour
through Finder.
Where to go next
- New users: read Projects — they're how TRMNLY organises your work — plus Worktrees for parallel git checkouts.
- Coming from another terminal: skim Keybinds for the deltas.
- Here for the agent: jump to Agent and Agent safety.