Voice Prompting Synara on Mac
By Kiran Jd
Published
To dictate into Synara on Mac, press Cmd+L to focus the composer, speak your task with your Speakmac shortcut, and review the text before you send it. Synara also has its own voice note button, but it needs a Codex account signed in with ChatGPT. A system-wide dictation app works whichever agent you pick.
Synara is a free, open-source desktop app for running coding agents such as Claude Code, Codex, OpenCode, Cursor and Pi in one workspace. It is not the AI security-camera company of the same name. Synara does not include a model: the agent you select does the work, using the account you already have.
Where does the prompt go in Synara?
Add your project, start a new task with Cmd+N, and click the composer. On a new chat it reads Do anything in Synara; while an agent is working it reads Ask for follow-up changes. Cmd+L focuses the composer from anywhere outside the terminal.
Before sending, check two things in the composer area: the provider and model (Cmd+Shift+M opens the picker), and whether the task runs in your local checkout or its own Git worktree. If several tasks are running on the same repository, a good prompt sent to the wrong environment still causes trouble.
How do you dictate a task into the composer?
- Install Speakmac and complete microphone and Accessibility setup. Choose a dictation model that works on your Mac.
- Install and sign in to at least one agent runtime, then confirm Synara detects it. Synara's installation guide covers this.
- Press Cmd+L so the composer has focus. Use your Speakmac Push to Talk or Hands-Free shortcut.
- Say the outcome, the scope, the constraints and how you will check the result. Stop dictating before you start reviewing.
- Fix file names and identifiers, attach files or a screenshot, and send only when the brief is ready.
Your Speakmac shortcut is set in Speakmac, not in Synara. If you turn on Synara's AppSnap, which captures a window when you press both Option keys, make sure your dictation shortcut does not use the same keys. See shortcuts and recording help.
What is a useful voice prompt?
Synara's first-task guide recommends four parts: outcome, scope, constraints and verification. That structure is easy to say out loud:
Outcome: the settings page should show an empty state when there are no saved profiles. Scope: only the profiles list and its tests. Constraints: reuse the existing card component and do not change the API. Verification: run the profiles tests and tell me the exact commands.
For a bug, describe what you saw and what should happen instead:
After I switch accounts, the old avatar stays in the header until I reload. Make the header update with the account. Keep the current loading state. Add a test that switches accounts twice.
If the change is larger, type /plan first. Plan mode makes the agent propose an approach before it edits anything. These are example briefs, not claims that any agent completed or tested the work. Read the diff and run the checks yourself.
What should you type instead?
| Speak the explanation | Type or attach the exact evidence |
|---|---|
| What you expected and what happened | Error output and stack traces |
| Scope and what must stay untouched | File paths, symbols and commands |
| Edge cases and UI behaviour | Screenshots, logs and line comments from the file preview |
Synara lets you comment on specific lines in a file preview and send those comments with your next prompt. That is often better than reading a line number aloud.
Do not dictate credentials or private customer data into an agent prompt. Speakmac transcribes on your Mac, but once you send, the selected provider receives the prompt and its policy applies. Synara says it stores workspace data locally and does not proxy normal provider traffic. Speakmac's privacy page covers its own transcription step.
What if text does not appear?
Press Cmd+L again and try a short sentence. If Speakmac produced a transcript but it did not reach the composer, use Copy Last Transcript in Speakmac and paste it in. If no transcript was produced, check the selected microphone, model and Accessibility permission in the Help Center. Those problems are separate from Synara's composer losing focus.
Does Synara have its own voice feature?
Yes. When it is available, the composer shows a Record voice note microphone button. A setting called Enter while dictating controls whether Enter stops and transcribes into the composer (the default) or stops and sends. You can also record while an agent is still working.
The limits, according to Synara's source code: voice notes are sent to ChatGPT's transcription service using the ChatGPT login of a Codex account in Synara, even if you are prompting Claude or another agent. A Codex account signed in with an API key does not unlock it, and each note can be up to 120 seconds.
If you already use Codex with a ChatGPT login and only dictate inside Synara, the built-in button may be all you need. A separate app like Speakmac helps when you want the same shortcut in Synara, your terminal, your editor and the GitHub review page, custom words for project names, or transcription that stays on your Mac.
Try Speakmac for Mac, or compare the workflow with voice prompting T3 Code and voice input for OpenCode.