Dictate your first useful draft
Start withA focused text field and a short message you can check.
Work towardA reviewed draft, with a recovery path if text lands in the wrong place.
Cue learning paths
Start with the task you need to finish, not a model name. These guides show what to bring, what to ask Cue, and how to check the result before you use it.
Use Dictation when you already know what you want to say. Speak a rough draft into a focused text field, then review names, numbers and commitments. Dictation is not the same as asking an Agent to act.
Start withA focused text field and a short message you can check.
Work towardA reviewed draft, with a recovery path if text lands in the wrong place.
Start withWho you are writing to, the agreed facts and a clear next step.
Work towardAn email draft that does not invent promises or send itself.
Start withCompleted work, blockers, next steps and unresolved requests.
Work towardA concise update that keeps partial progress separate from completion.
Start withA technical note with names, identifiers and numbers to preserve.
Work towardA checked brief with exact terms, a clear request and unresolved questions.
Start withConfigured Cue Dictation, an editable draft, and shareable ticket facts and company rules.
Work towarda useful support reply and internal checklist without invented refunds or repair deadlines.
A transcript is source material, not a decision log. Start with notes or a recording you are allowed to use. Correct the relevant excerpt, separate facts from suggestions, and carry only the necessary context into the next task.
Start withA reviewed meeting excerpt with decisions and open questions.
Work towardAn action table whose owners and deadlines are supported by the source.
Start withPermission to use the notes and a few relevant, corrected excerpts.
Work towardA requirements draft that separates evidence, proposals and unknowns.
Start withA permitted recording, source playback and a clear next task.
Work towardA focused context packet with source references and unresolved questions.
Start withReviewed meeting notes, the next meeting's goal and its time limit.
Work towardA timed agenda draft, a pre-read checklist and unanswered questions.
Start withTwo reviewed meeting excerpts, their dates, a known approval rule, and configured Cue Agent.
Work towarda source-linked decision change log and a checklist of affected actions to inspect.
For analysis or action, use an Agent with an explicit goal, selected context, constraints and a way to verify success. Bringing an Agent into your current app does not mean it can see every file or act without the required permissions.
Start withOne clear goal and the smallest useful piece of app context.
Work towardA result you can inspect against acceptance checks before taking the next step.
Start withA task brief, relevant evidence and access to your chosen coding Agent.
Work towardAn explicit handoff, with connection, context and execution checked separately.
Start withSteps, expected behavior, observed behavior and non-sensitive evidence.
Work towardA bug report another person or Agent can reproduce, not an assumed diagnosis.
Start withSource notes that disagree about scope, status or decisions.
Work towardA brief that preserves source authority and asks about unresolved conflicts.
Start withA reproducible bug in an authorized project and authenticated Claude Code access.
Work towardA scoped patch with a failing regression test and passing results after the fix.
Start withAn identified Git change, its intended behavior and Codex access.
Work towardPrioritized findings with evidence and a clear decision on the next fix.
Start withA non-sensitive practice CSV and Python 3 to run the checks.
Work towardA checked subtotal with included, excluded and unresolved rows.
Start withConfigured Cue Dictation, a scratch note, and authorized access to the interrupted task's evidence.
Work towarda checked recovery brief that preserves completed work and retries only the verified remaining step.
Start withA passage already open in the app you are working in, and one small output you need from it.
Work towardA checked result whose source you proved, plus a clear boundary for actions that leave your machine.
Do not pass the entire conversation along by default. At each handoff, keep the evidence needed for the next task and retain a link or source identifier so the result can be checked.
Extract actions and leave missing owners or dates explicitly unassigned.
Use those checked facts to describe progress and blockers.
Draft a follow-up with one clear request; review it before sending.
Turn a small interview excerpt into evidence-linked requirements.
Resolve disagreements with approved scope before assigning work.
Give an Agent one bounded task and explicit acceptance checks.
Capture the reproduction steps while the relevant app is in front of you.
Prepare the brief and explicitly supply it to the connected coding Agent.
Compare the proposed fix and test evidence with your original report. A generated patch is not proof that a bug is fixed.
Choose the capture workflow first, then evaluate the result. These additional resources are currently in English.
Use the benchmark methodology and linked public resources to define useful checks. A framework for evaluation is not a claim that Cue outperforms other products.