Voiskey and Jev: Voice Commands That Do More Than Dictate

A practical method for separating dictated text from device actions

Voiskey Voiskey Team
Voice CommandsDevice AutomationVoiskey Jev Voice

Two Outputs From One Microphone

Voiskey and Jev share one microphone but produce two different outcomes, and separating them is the first practical step. Dictation turns speech into polished text at the cursor; a device command asks the agent to act. Before teaching any routine, list which sentences belong to each category. A request such as summarize this paragraph is text work, while open the folder named drafts is an action request. Write the two lists side by side and review them with the person who will speak them aloud. Overlap at this stage turns into a wrong action later, so settle ambiguous wording before you rehearse. Keep both lists visible during the first week.

Map the work people already do by keyboard before adding voice. Sit beside one colleague for half an hour and note every switch between typing, clicking, and waiting. Mark the tasks that are short, repetitive, and easy to describe in one sentence; those are the best candidates for a spoken command. Mark the tasks that need judgment, negotiation, or careful reading; keep those on the keyboard for now. Ask the colleague which interruption costs the most attention and why. The answer usually names a single bottleneck rather than a broad category. Record the observation in plain language, with the time of day and the app involved, so the next session can revisit the same situation with the same app open.

Define what a correct command looks like before anyone practices. For each candidate task, write three parts: the sentence a person says, the action the agent should take, and the visible result that proves it happened. Add a fourth line for what should happen when the sentence is misunderstood. These four lines become your acceptance sheet, and they matter more than a long list of features. Ask two people to read the sheet and predict the outcome of each command independently. When their predictions differ, the wording is still unclear. A command that cannot be described this precisely is not ready for a live session, no matter how well it worked once in a demonstration.

Set a scope boundary for the first month. Choose one app, one folder, and one type of task, then resist the urge to generalize early. Name the person who approves changes to the command list and the person who performs the work. Agree on how a command can be withdrawn if it causes confusion. Keep a written note of that decision rather than relying on conversation. If a new request arrives during the month, park it in a list for the next review instead of adding it immediately. Narrow scope is not a lack of ambition; it is the condition that makes evidence trustworthy when you later widen the routine.

Phrasing Commands the Agent Can Act On

Start every command with a verb. Open, move, rename, create, close, and switch are clearer than phrases that begin with context or politeness. Keep one action per sentence and stop when the action is complete. If a task needs two steps, say them as two sentences with a short pause between. This rhythm gives the agent a clean boundary and gives you a chance to correct the first step before the second begins. Speak at a normal pace rather than exaggerating each word. Practicing verb-first sentences for ten minutes before a live session often removes more errors than any change to the tool itself, because the habit survives when the phrase sheet is not in front of you.

Name objects exactly as they appear on screen. Say the folder name, the file name, and the path in full rather than pointing with words like that one or the thing from yesterday. If a name is long or unfamiliar, add it to the vocabulary you are teaching so it comes through correctly. Distinguish similar names deliberately, for example by including the parent folder. When two files share a prefix, say the distinguishing word first so the agent does not have to guess. Write the exact names in your phrase sheet and rehearse them aloud twice before the first live attempt, then confirm each one appeared correctly before adding the next command.

Remove stacked conditions from spoken commands. A sentence that contains if, unless, and after in the same breath asks the agent to resolve three decisions at once. Split the request into a sequence: first check the state, then act, then confirm. Use numbers when order matters and say the number aloud before each step. Avoid vague modifiers such as quickly, properly, or a few; they carry no instruction the agent can follow. Keep the tone neutral and direct. Short declarative sentences read well to a listener, behave predictably when spoken into a microphone, and give the operator an obvious place to stop and check the screen before moving on.

Build a phrase sheet of ten commands and test it by reading it aloud. Include the sentence, the expected action, and one likely mishearing. Ask a colleague to listen without seeing the sheet and repeat what they heard; gaps in their repetition reveal unclear words. Revise any command that needs explanation. Put the sheet where the operator can see it during practice, then remove it once the phrases feel natural. Review the sheet after the first week and delete commands nobody used. A short list that people actually say beats a long list that looks complete, and the sheet becomes more accurate every time someone drops a phrase that never felt natural. Bring the sheet to every practice session and mark the commands that still need a pause between steps.

Supervised Sessions Before Unattended Use

Run the first sessions with a spotter who watches the screen while the operator speaks. The spotter confirms that the intended action matches the visible result and writes down every mismatch. Keep the operator's hands free, but keep a manual fallback ready in case a command produces nothing. Agree on a single word that stops the session immediately; both people should use it without hesitation. Begin with the three easiest commands and repeat them until they are boring. Boring is the goal at this stage, because predictable repetition exposes small errors that a rushed demonstration hides, and it shows whether the operator can stay calm when a command does nothing.

Measure three numbers in each session: commands attempted, commands completed without correction, and time spent recovering. Write them on a shared sheet immediately after the session, not from memory the next day. Add a short note for every recovery describing what was said and what happened. Do not average away a rare but serious misfire; list it separately and decide whether the command stays in scope. Compare the totals across sessions with the same task set. If the numbers move because the tasks changed, say so instead of presenting the change as progress, and keep the previous session totals beside the new ones.

When a command misfires, stop and restate the current state before trying again. Say what you believe is on screen, then ask the spotter to confirm. Repeating the same sentence louder rarely helps; change one element, such as the object name or the verb. If the second attempt also fails, switch to the manual path and note the exact wording that failed. Do not chain several guesses together, because each guess changes the state and makes the original problem harder to diagnose. A calm restatement is faster than an improvised sequence of corrections, and it leaves a clean description that the next person can reproduce.

Close every session with a five-minute review. Read the notes aloud, mark each item as wording, environment, or expectation, and assign one owner per item. Decide whether the next session repeats the same tasks or adds one new command. Keep the old notes attached to the command sheet so a later change can be compared with the original behavior. Announce the next session date and who will attend. The review is not paperwork; it is the moment where observation becomes a decision, and skipping it turns a week of practice into anecdotes that nobody can act on later. Keep the review under ten minutes so the team will actually hold it every week.

Installing the Command Layer

Install Voiskey on your Mac or Windows device from the official page, then complete the first run with the shortcut you will actually use. Start with one day of dictation only, no device commands. Type nothing for routine messages; speak, let the app remove filler words and false starts, then edit the result. This single day teaches you how text appears at the insertion point and how corrections feel. Keep a note of phrases that came through badly and add the names or acronyms to your vocabulary. The download takes minutes, but the calibration is the part that deserves attention, so resist the urge to judge the whole setup on the first afternoon. Say a few long sentences with numbers and proper names, then compare what appeared with what you meant to say.

Use the second day to connect speech and action. Voiskey turns spoken requests into clean instructions, including file names, paths, and commands, which is exactly what a device agent needs. Say the request as you would to a careful colleague, then check that the polished instruction still means what you intended. Polishing can shorten a sentence, so read the result before the action runs. Practice five commands in a safe folder where a wrong move costs nothing. If a command feels risky, keep it out of the spoken set until the phrasing is stable and a recovery path has been written down for the person who will say it. Keep the safe folder small enough that listing its contents takes one second.

Plan for the early stage of the integration. Direct voice commands that guide the agent through everyday device tasks are still under active refinement, so behavior can change between sessions. Keep a manual alternative for every command on your list and know how to reach it without thinking. Test the same command at the start and end of each week and note any difference. Report persistent mismatches with the exact words you used. Treat the integration as a promising assistant rather than a finished operator, and your expectations will match what it can currently do during a normal working week. When a command works after a change, write down the date so a later regression is easier to spot.

Set a personal rule for when speech is the right input. Use it when your hands are occupied, when the sentence is short, and when a mistake is easy to see and undo. Switch back to the keyboard for long edits, precise selection, and anything you cannot verify at a glance. Write these three rules on the same sheet as your command list. Review the rules after two weeks and adjust them based on what actually happened. A clear split between spoken and typed work prevents frustration, keeps the routine sustainable, and makes it obvious which failures belong to speech and which belong to the agent. Revisit the rules whenever a new app joins the routine, because its text fields may behave differently from the ones you practiced in.

Extending the Routine Safely

Add one command per week, never more. Write the new sentence, the expected action, and the recovery step before you try it. Run it ten times in a safe setting and record the result each time. If it fails twice in a row, remove it and revisit the wording next week. This slow cadence feels unnecessary when everything works, but it protects the trusted commands from changes that arrive together. Keep a single changelog with dates so you can tell which version of the phrase sheet produced which outcome. Bring the changelog to every Friday review and read the last two entries aloud before deciding anything.

Review the command list every Friday with the person who uses it most. Delete anything unused for two weeks; a long list invites accidental matches and makes practice feel like a chore. Rewrite any command that needed an explanation during the week. Check that the names in the list still match the folders and files on the device, because renaming a folder quietly breaks a command that used to work. Note the review in the changelog with the date and the person who approved the change. If a deletion feels risky, keep the old phrase in a retired section for one month rather than erasing it.

Write a handover note for anyone who takes over the routine. Include the shortcut, the phrase sheet, the safe practice folder, the recovery steps, and the current known limits. Ask a colleague to follow the note without your help and mark every point where they hesitate. Fix those points in the note rather than explaining them verbally. Keep the note next to the device where the work happens. Handover quality is measured by whether someone else can run the routine for a day without calling you. Include one worked example with the exact sentence, the expected screen, and the recovery step for a failed run.

Decide at the end of the month whether to widen the routine. Use the changelog, the weekly measurements, and the recovery notes as evidence rather than impressions. If most commands work and recoveries are rare, add one new app or task type. If failures cluster around a single phrase, fix that phrase before expanding. Download Voiskey on any additional device only after the routine is stable on the first one, and repeat the calibration there. Expansion should follow evidence, not enthusiasm, and the written record tells you which one is driving. Set the next review date before you close the current one, and name the person who will bring the measurements.

FAQ s

Voiskey is a system-level voice typing app for desktop computers. You press a shortcut, speak, and text appears at the insertion point in any app. It removes filler words and false starts, and it turns spoken requests into clean instructions, including file names, paths, and commands, so a device agent receives a usable sentence instead of a rough transcript.

Speak the request the way you would to a careful colleague. Voiskey removes stutters and self-corrections, adapts punctuation and tone to the task, and turns the spoken request into an AI-ready prompt that can carry file names, paths, and commands. Read the polished sentence once before it runs and correct any term that changed the meaning.

Voiskey learns names, jargon, acronyms, and spelling preferences. Add the unusual product name or short form during the first sessions, then say the command once in a practice folder and confirm the wording that appears. Once the term is recognized, the same command can be spoken with the exact name rather than a vague description.

Voiskey works at system level, so you press a shortcut in whatever app is in front of you and speak. It removes false starts, applies the punctuation and tone the task needs, and produces a clean instruction with file names, paths, or commands. That polished sentence is what the device agent can act on, and you can edit it before anything runs.

Say less. Get more
Say less. Get more
Say less. Get more
Say less. Get more
VoiskeyVoiskey

Your voice, amplified.

Download Voiskey for Mac and Windows to practice dictation and device commands together.

Use it across macOS, Windows, iOS, and Android

Use it across