Say what you mean, then check it against the code.
You often understand a problem well before you have typed it up. Cue is a desktop voice agent for Mac and Windows that captures that explanation in the field you focused, so you can review it and hand a bounded task to the coding agent you already use. Each workflow below links to a published guide, most of them with a practice exercise.
Six workflows, from a spoken note to a checked change
The guides use fictional or disposable projects, and their sample outputs are illustrative rather than recorded Cue runs. Follow the controls in your installed version.
Voice coding in your editor
- Start with
- The relevant code, selected or visible, in a supported desktop editor.
- Ask Cue for
- A proposed change to the code, or an explanation of it.
- Check before you share
- Every changed line and your normal tests. Nothing is applied for you.
Technical terms and identifiers
- Start with
- A scratch note and an authoritative reference for the exact names.
- Ask Cue for
- A spoken explanation, with identifiers copied rather than guessed.
- Check before you share
- Names, values and negations match the reference, since one missing "not" changes the task.
Spoken bug report to a brief
- Start with
- A concrete bug, with the affected app visible.
- Ask Cue for
- Steps to reproduce, expected and observed behavior, known conditions and acceptance checks.
- Check before you share
- The brief describes what you saw, without a guessed diagnosis, and goes to the correct repository.
A small fix with Claude Code
- Start with
- A reproducible bug in a disposable or authorized project, and an authenticated Claude Code.
- Ask Cue for
- A captured symptom you hand over, with an agreed failing example and scope.
- Check before you share
- The regression test failed first, the diff stays in scope, and release approval stays separate.
A review with Codex
- Start with
- A clearly identified change, its intended behavior and Codex access.
- Ask Cue for
- A spoken review goal for a read-only review of the exact diff.
- Check before you share
- Each finding holds up against the changed lines, and approving a fix stays a separate step.
Release notes from a diff
- Start with
- A selected diff and the record of what was actually tested.
- Ask Cue for
- Draft wording from your walkthrough, one statement per change.
- Check before you share
- Every sentence traces to the diff and the test record, and you publish the approved text yourself.
Three things to keep separate
Dictation
It turns your speech into text in the field you focused. The default shortcut is Option on Mac and Right Alt on Windows, and your installed settings take precedence.
An Agent
It works on a request. Seeing Claude or Codex in the agent selector does not mean it is connected. An external agent keeps its own login, permissions, models and billing.
Context
Context is what you hand over: the brief and the material you selected. Cue does not promise to share your full history, hidden context or permissions across agents.
Use Cue voice with Claude Code, Codex and your task context explains how these parts fit together.
What Cue does not do here
- It offers no public programmatic surface. Cue does not document a public API, SDK, CLI or MCP server. The developer guide keeps the current status.
- It does not verify code for you. A confident explanation is not a test result. Read the diff and run the tests yourself.
- It does not set your data policy. Requests can send selected text and other context to the provider handling them. Read the Privacy Policy and follow your organization's rules for private code.
- It does not prove a result. These pages describe methods. They contain no productivity figures or customer results.
Cue for engineers FAQ
Does Cue replace Claude Code or Codex?
No. Cue helps you capture the problem and write the brief. Claude Code and Codex run under their own login, permissions, models and billing, and you hand the brief over explicitly.
Does Cue have a CLI, API or MCP server?
Cue does not document a public API, SDK, CLI or MCP server for third-party integration at this time. The workflows on this page run through the Cue desktop app.
Will Cue spell my code identifiers correctly?
Do not assume so. Speak the explanation, copy the exact identifier from an authoritative source, and do not ask a language model to guess its spelling. The technical terms tutorial shows how to check names, values and negations.
Which editors can I dictate into?
Cue dictates into supported text fields in desktop editors such as Cursor. Insertion depends on the editor, the field and your installed version, so check the result where you work.
What happens to the code I select?
To carry out a request, Cue may send selected text, clipboard content, prompts and relevant thread context to the AI provider handling it. Check your organization's rules before you use Cue with private code, and read the Privacy Policy.