All pages
Lesson 6.2 · The browser and your desktop · 10 min
Screenshots, voice and images
Send screenshots to the agent, dictate, and talk hands-free.
Show the agent what you are looking at instead of describing it, and talk to TRMNLY instead of typing. This lesson covers sending screenshots and images to the agent or to a coding CLI in a pane, dictating into the terminal, and holding a spoken conversation hands-free.
Time: 10 minutes · Works on: macOS app
Send a screenshot to the agent
TRMNLY sets up two global shortcuts on first launch. Global means they work while another app is in front.
| Shortcut | What it does |
|---|---|
⌃⌘S |
Screenshot to Chat: the capture lands in the agent's message box, ready for you to add a question |
⌃⇧⌘S |
Screenshot to Terminal: the image's file path is typed into the focused terminal pane |
Try the first one:
- Open a page or an app with something you want to ask about, such as an error dialog.
- Press
⌃⌘S. The pointer becomes a crosshair. - Drag out the area you want. Press Space first to capture a whole window instead, or Escape to cancel.
- The screenshot appears in the agent's message box. Type a question, for example "What does this error mean?", and send it.
You can also click the camera button beside the agent's message box. Its tooltip reads Capture a screenshot into the chat.
macOS may ask you to allow screen recording the first time. Grant it in System Settings → Privacy & Security.
Send a screenshot to a CLI agent in a pane
If you run Claude Code, Codex or another coding CLI in a terminal pane, ⌃⇧⌘S is the one to use.
- Click into the pane running the CLI.
- Press
⌃⇧⌘Sand select an area. - The image is saved inside your project, in
.trmnly/attachments/, so the CLI can read it. Its path is typed at the prompt. A CLI agent gets it as@/path/to/image.png. A plain shell gets an escaped path. - Nothing is submitted. Add your question and press Return.
You can do the same from a pane's right-click menu with Screenshot for Agent….
To change either shortcut, or to add a key that attaches a file instead, open Settings → Shortcuts.
Paste or drag images into the agent
The agent's message box accepts images directly:
- Paste an image you copied, with
⌘V. - Drag an image file from Finder onto the message box.
The image goes with your next message. All of the agent's model providers accept images. A very large image is scaled down rather than dropped.
Dictate into the terminal
Dictation types what you say into the focused pane as plain text. It doesn't run anything and doesn't press Return for you.
- Click into a terminal pane.
- Press
⌘I, or choose View → Dictate into Terminal. - Speak. A small floating window shows the transcript as you talk and your input level. It doesn't take keyboard focus away from the terminal.
- Press
⌘Iagain to stop. The text is inserted at the prompt. - Read it, fix anything that was misheard, and press Return yourself.
The mic button in the right sidebar's header does the same thing, with two gestures. Tap it to start and tap again to stop. Or hold it down and talk, and the text is inserted when you let go.
The text goes into the pane you were in when you started, even if focus moves while you speak. It is inserted all at once when you stop, because the recognizer keeps revising what it heard.
The first time, macOS asks for Microphone and Speech Recognition permission. Recognition runs on your Mac. Nothing listens until you start, and the mic closes when you stop.
Talk to the agent hands-free
Hands-free mode is a spoken back-and-forth. It listens, sends what you said, waits for the reply, reads it aloud, and listens again.
- Press
⌃⌘I, or choose View → Hands-Free Conversation. - Say what you want. When you pause for about a second, it sends.
- A floating panel shows the phase: listening, thinking, or speaking.
- Listen to the reply. Then speak again.
- Press
⌃⌘Iagain to stop.
The waveform button in the agent's message box controls the same mode. Its menu chooses who you are talking to:
- Hands-free with the TRMNLY agent: your words become a message in the agent sidebar, and the agent's reply is read aloud.
- Hands-free with the focused pane: your words are typed into the pane you were in and submitted with Return. Use this to talk to Claude Code or another CLI. The "reply" is whatever the pane prints once it settles.
The mic is closed while a reply is being read, so the app doesn't hear itself. To interrupt, press ⌃⌘I or use the button on the floating panel. Stop hands-free in the button's menu ends the session.
Choose the voice
Open Settings → AI Agent and find Voice & speech:
- Speak assistant replies reads the agent's answers aloud outside hands-free mode too.
- TTS backend picks System voice (offline) or ElevenLabs. ElevenLabs needs its key in Settings → API Keys.
- Transcription backend picks Apple (on-device) or Whisper (OpenAI-compatible).
- Wake word listening ("hey trmnly") lets you start the agent's voice input by saying its name. This is separate from hands-free mode and off unless you turn it on.
Check yourself
⌃⌘Sputs a screenshot in the agent's message box from any app.⌃⇧⌘Stypes an@path into a pane running a CLI agent.⌘Iinserts your spoken words at a shell prompt without running them.- One hands-free exchange with the agent works start to finish with no typing.
Go further
- Custom shortcuts: the screenshot and attach-file actions, and how to rebind them
- Dictation: how dictation targets a pane
- Hands-free: both targets and how interruption works
- Custom shortcuts lesson: bind your own key chords