1. Read Planned as a roadmap state—not a shipped control
Zed’s current roadmap places “Dictation for Assistant (inline and chat-based)” under Planned. It separately lists “Voice Input with Voice Activation to Chat in Zed” in the open triage section.
That is evidence of product direction, not a current setup procedure. Recheck the live documentation and release before claiming a shortcut, button, processor, language list, or availability rule for a feature that has not moved into shipped docs.
2. Recognize the existing microphone as collaboration voice chat
Zed Channels support voice chat alongside shared projects and collaborative notes. The channel microphone mutes or unmutes call audio so collaborators can hear one another.
Zed says joining a channel automatically shares the microphone with other users in the call when the operating system permits it, unless mute-on-join is configured. That microphone carries call audio; the documentation does not describe it as turning speech into text in an editor or agent field.
3. Use a current Mac text-input path for Zed today
Click the intended Zed field until the insertion point is visible, then trigger the shortcut shown under System Settings → Keyboard → Dictation. Apple says Dictation can enter text anywhere you can type.
Test a short unsent phrase before a long prompt. Editor buffers, agent panels, inline assistants, terminals, and embedded controls can expose focus differently, so verify where the text actually lands and keep the clipboard available as a fallback.
- Confirm the active project, pane, and field.
- Dictate a short reversible draft.
- Stop listening and inspect the full transcript before submission or execution.
4. Keep call privacy, AI privacy, and dictation privacy separate
Zed’s current privacy policy says it processes but does not store audio, video, or screenshare content from collaboration sessions. That statement belongs to collaboration calls.
It does not automatically describe future assistant dictation, macOS Dictation, a third-party input tool, or text later sent to an AI provider. For each path, identify who receives microphone audio, who receives transcript text, what context joins the request, what is retained, and which account or opt-in control applies.
5. Review agent prompts and terminal text before acting
A prose transcript in a Zed agent field can become an instruction to a model with project context. A transcript in the integrated terminal can become an executable command when Return is pressed. Neither destination should receive an unchecked guess.
Use voice for intent, explanation, comments, and reviewable requirements. Type or carefully verify exact source syntax, shell operators, file paths, package names, secrets, flags, and commands.
6. Add IraVoice only for the compatible local Mac workflow
IraVoice 0.7.1 is a separate option for macOS 26+ Apple-silicon Macs when one held-key habit should create editable text across supported Zed, browser, email, document, and developer fields. Speech recognition and supported formatting run on the Mac after model setup.
IraVoice does not add a Zed extension, control collaboration calls, read a project, dictate reliable source syntax, choose an agent, press Send, or execute terminal commands. Voice-to-Spec can instead create reviewable Markdown that you manually paste into the intended Zed surface.