Voice Prompting Conductor on Mac
By Kiran Jd
Published
To use Speakmac with Conductor on Mac, click the message box in a workspace chat, or press ⌘N to start a new workspace, then dictate with your Speakmac shortcut and read the brief before you send it. Conductor also has its own experimental Voice mode on macOS 26. Whichever you use, check file names and scope before the agent starts.
Conductor is a Mac app from Melty Labs that runs Claude Code, Codex, Cursor and OpenCode agents in parallel. Each task gets its own workspace with a branch, working tree, terminal and diff. It is not the Netflix Conductor workflow engine (now Conductor OSS and Orkes), the Conductor SEO platform, or Retell's Conductor assistant. Speakmac only supplies the text. The agent you pick in Conductor does the work after you send.
Where does the prompt go in Conductor?
On the Mac there are three places where you write to an agent:
- New workspace.
⌘Nopens the new-workspace composer. The prompt you write there is sent when the workspace is created, so check the repository and starting branch first. - Workspace chat. The message box at the bottom of each chat tab. Separate tabs can run different agents in the same workspace.
- Diff comments. Comments on changed lines in the Diff Viewer (
⌘⇧D) become composer attachments, so the agent gets line-level feedback.
If you would rather work in a terminal, Conductor's experimental Big Terminal Mode runs claude, codex or another CLI directly. For that route, see our guide to dictation in Claude Code.
How do you dictate a task into Conductor?
- Install Speakmac and complete microphone and Accessibility setup. Choose a dictation model that works on your Mac.
- Open Conductor and follow its install and setup checks. It uses the Claude Code, Codex or Cursor credentials already on your Mac.
- Open the workspace and chat tab you mean to use, and check the agent and model in the model picker.
- Click the message box so it has keyboard focus, then use your Speakmac Push to Talk or Hands-Free shortcut. Say the outcome, the current behaviour, the constraints and how to check the result.
- Stop dictation, type
@to mention files or folders, correct any identifiers, and send.
Conductor lets you choose whether Return or ⌘Return sends, and in current versions a message sent while an agent is working steers that running turn. Leave Speakmac's Send After Dictation shortcut unused until you trust your transcripts: it presses Return, which may send an unchecked brief or redirect an agent mid-task.
Speakmac's shortcut is configured in Speakmac, not in Conductor. See shortcuts and recording help if it clashes with another app.
What is a useful voice prompt?
Give each independent task its own workspace, and say what done looks like. A new-workspace brief:
Add rate limiting to the public search endpoint. Keep the current response shape, return a 429 with a Retry-After header, and make the limit configurable. Add a test that makes eleven requests and checks that the eleventh is rejected.
A follow-up in the same chat after you review the diff:
The empty state hides the Add task button on narrow screens. Keep the copy, move the button above the illustration, and check the layout at 375 pixels wide. Do not change the task list itself.
For broad or risky work, type /plan before you dictate so the agent proposes an approach before editing files. Conductor's agent modes page suggests Plan Mode for migrations, unclear refactors and bugs with an unknown cause. Rules you would otherwise repeat every time belong in AGENTS.md, CLAUDE.md or repository settings, not in each spoken prompt.
These are example briefs, not claims that an agent completed or tested the work. Review 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, attached with @terminal |
| UI behaviour and edge cases | File paths and symbols, added with @ mentions |
| Scope and what must not change | Screenshots, specs and logs as attachments |
Do not dictate credentials or private customer material into an agent prompt. Local transcription does not make the agent request local: after you send, the selected provider's policy applies, and chats in Conductor's cloud workspaces are stored on Conductor's servers. Speakmac's privacy explanation covers its own transcription step.
What if text does not appear?
Click back into the message box and try a short sentence. If Speakmac produced a transcript but it did not land, use Paste Last Transcript, or Copy Last Transcript and paste it yourself.
If your Speakmac shortcut triggers a Conductor action instead, pick a different shortcut; Conductor uses many key combinations, listed in its in-app shortcuts dialog. Update Conductor as well: earlier releases fixed bugs where it could block global keyboard shortcuts (0.81) and Option-key input in its text fields (0.54). If no transcript was produced, check the microphone, model and Accessibility permission in the Help Center.
Does Conductor have its own voice feature?
Yes. The composer already accepted voice messages by mid-2026, and Conductor 0.91.0 (15 October 2026) added Voice mode under Settings → Experimental. On macOS 26 it transcribes on your Mac and shows your words in the message box as you speak, with a live microphone level. We have not tested it, and we have not confirmed its languages, custom vocabulary or behaviour on older macOS versions.
If you dictate only into Conductor and you are on macOS 26, try Voice mode first. A system-wide app helps when you also dictate into a terminal, GitHub reviews or Linear, want the same shortcut everywhere, or need custom words for repository names and identifiers. Conductor's co-founder Charlie Holtz described the same pattern in June 2026: he presses ⌘N, dictates through a separate local speech-to-text app and presses Return.
Try Speakmac for Mac, or compare the workflow with voice prompting in T3 Code and voice prompting in Codex.