DictaFlow

New user guide

Getting started with DictaFlow

A practical guide to getting DictaFlow typing where you already work.

If you only remember one thing: hold your trigger, speak naturally, release, and DictaFlow will type the cleaned-up text wherever your cursor is.

Quick start

  1. 1Install DictaFlow and sign in.
  2. 2Open any app with a text cursor.
  3. 3Hold your trigger, speak, then release.
  4. 4Set shortcuts, model choice, and custom instructions in Settings.

Watch first

Two short tutorials cover the basics.

These are the same quick-start videos used inside the desktop app.

1-Minute Beginner Tutorial

Sign in, start recording, release to type, and find the basic settings.

Open on YouTube →

1-Minute Advanced Tutorial

Smart Edit, work domains, Typing Mode, and power-user settings.

Open on YouTube →

Step 1

Try your first dictation

Open any app with a text cursor: Gmail, Outlook, Slack, Word, Notes, VS Code, a browser, or a web form. Hold your dictation trigger, say one or two sentences, then release. The text should appear where your cursor was.

Holdstart recording
Speaktalk normally
Releasetext appears

Step 2

Set the shortcuts you will actually use

DictaFlow has separate triggers for normal dictation, AI Smart Edit, and Typing Mode. To change any shortcut, open Settings → Shortcuts & Interface, click Change Shortcut, then press the new key combo. If that shortcut is already used by another DictaFlow mode, the app will warn you.

Dictation

Your main hold-to-talk trigger. On Windows, this starts as Ctrl + Win. You can also use a USB dictation foot pedal as the trigger.

AI Smart Edit

Use this when you want cleanup, rewriting, formatting, or selected-text edits. On Windows, this starts as Alt + Win, but you may need to turn it on first.

Typing Mode

Use this when normal paste fails in remote or locked-down apps. On Windows, this starts as Ctrl + Alt + Space, and needs to be enabled before use.

This guide is showing Windows shortcuts. On Mac, the defaults are Fn, Control + Fn, and Control + Option + Space.

Step 3

Use AI Smart Edit when you want more than a transcript

Use normal dictation for quick text. Use AI Smart Edit when you want DictaFlow to clean up, rewrite, generate, or follow an instruction while you talk.

Correct yourself: say “actually” or “I mean” and keep going.
Format as you speak: “make this a list” or “format this as Python.”
Edit selected text: highlight a paragraph and say “make this clearer.”
Generate drafts: “draft a polite follow-up email about the budget.”

Desktop context

How app-aware formatting works

On Windows and Mac, DictaFlow can use the foreground app and window as context when building a cloud-assisted dictation or AI Smart Edit request. It groups common desktop apps into practical categories, then adds formatting guidance for that category.

Email and office apps

In Gmail, Outlook, Mail, Word, and similar apps, DictaFlow can favor email structure or business-friendly formatting.

Messages and notes

In Slack, Teams, Messages, Notes, Notion, and similar apps, it can favor conversational or organized note formatting.

Code and terminals

In code editors and terminals, it can preserve technical names, commands, paths, and syntax more carefully.

AI and coding tools

In tools such as ChatGPT, Claude, Cursor, VS Code, or Codex, AI Smart Edit treats a prompt-like request as text to insert instead of answering it.

What it does not do: app-aware formatting does not control the app or guarantee one fixed format for every field. Your Work Domain, Knowledge Base, mode, and custom instructions still matter.

This context-aware cleanup is a desktop cloud feature. Local Offline standard dictation does not use cloud AI formatting.

Step 4

Teach DictaFlow your words

Open Knowledge Base. Use Custom Vocabulary for names, acronyms, drug names, product names, client names, and technical terms. Use Text Expansions for phrases or templates you repeat.

Custom Vocabulary

Words DictaFlow should spell exactly: “Aerius,” “DictaFlow,” client names, library names, drug names.

Text Expansions

Short voice triggers for repeated text, disclaimers, signatures, snippets, or templates.

Step 5

Choose the right transcription model

Open Settings and look for Model. This controls how DictaFlow balances speed, accuracy, and instruction-following.

Studio Session

The everyday choice. Use it when accuracy matters, especially for medical, technical, or name-heavy dictation.

Supersonic Sprint

The faster option. Good for quick notes, short replies, and rapid capture when speed matters more than polish.

Command Suite

The Pro transcription option. In our synthetic audio tests it preserved names, dates, amounts, and punctuation well, but it did not reliably create lists or headings by itself.

Step 6

Build custom instructions from rules you can check

On Windows, open Settings → Dictation Behavior → Prompt → Edit Custom Instructions, paste one prompt, then choose Save & Close. Under Settings → Shortcuts & Interface → AI Smart Edit, confirm AI Smart Edit is enabled, note or change its shortcut, and choose its Output behavior. The Windows default shortcut is Alt+Windows.

Selected text and new dictation work differently

  • Edit existing text: highlight it in the destination app, use AI Smart Edit, say the immediate edit request, then stop recording. Selected-text transformations remain available when voice-request generation is off.
  • Format new dictation: place the cursor with no text selected, use AI Smart Edit, and dictate the content. Your standing Custom Instructions can shape that content.
  • Create from a request: for a no-selection request such as “Draft an email…”, enable Settings → Show advanced options → Advanced Configuration → Generate content from voice requests (AI Smart Edit). It is off by default. Without it, requests and questions are generally preserved as text rather than carried out.

Treat what you say during AI Smart Edit as the immediate task and Custom Instructions as standing guidance. Keep them compatible. If they conflict, results can vary; rewrite or temporarily remove the standing rule instead of relying on one instruction to override the other. DictaFlow's safety and mode rules still apply.

Write the instruction in this order

  1. State the output you want: a bullet list, numbered procedure, or lightly cleaned transcript.
  2. Say what must stay exact: names, dates, IDs, commands, amounts, tone, or exclusions.
  3. Say what must not be added. Remove contradictory rules instead of assuming one will always win.

Concrete rules worked better in these bounded tests than broad adjectives. For example, say Start every item with '- ' instead of only saying “use bullets.”

AI Smart Edit · project updates

Action bullets with owners and deadlines

Format project updates as a plain bullet list. Start every item with '- '. Keep each owner, action, and deadline together. Preserve names and technical terms. Do not add tasks, dates, priorities, headings, or status labels.

Evidence: text-only production formatter call using the deployed prompt builder and model; ASR and HTTP were bypassed. Repeated three times with the same output.

Synthetic inputQuick project update. Maya Chen owns the OAuth 2.1 callback fix by Friday. Luis Romero will benchmark the Postgres 16 migration. I need to send the Acme North risk log before 4:30 p.m.

Actual formatter output- Maya Chen owns the OAuth 2.1 callback fix by Friday.
- Luis Romero will benchmark the Postgres 16 migration.
- I need to send the Acme North risk log before 4:30 p.m.

AI Smart Edit · technical notes

Numbered technical procedures

Format technical procedures as a numbered list. Preserve commands, identifiers, flags, versions, environments, and conditions exactly. Do not infer commands or add steps.

Evidence: text-only production formatter call using the deployed prompt builder and model; ASR and HTTP were bypassed. Repeated twice with the same output.

Synthetic inputSet ENABLE_V2_SYNC=false. Run pnpm test:integration. If the exit code is 0, deploy to staging only.

Actual formatter output (line breaks rendered)1. Set ENABLE_V2_SYNC=false.
2. Run pnpm test:integration.
3. If the exit code is 0, deploy to staging only.

AI Smart Edit · casual chat

Keep the correction, not both versions

Keep chat casual and brief. Remove filler. When I correct myself with 'actually,' delete the earlier wording and keep only the correction. Preserve greetings, names, times, and product codes. Do not add a sign-off, apology, emoji, or corporate language.

Evidence: text-only production formatter call using the deployed prompt builder and model; ASR and HTTP were bypassed. Repeated three times. All three removed the superseded time; two rendered 2:15 and one wrote “two fifteen.”

Synthetic inputHey Sam yeah the beta looks good I can join at two actually two fifteen and I'll bring the Kestrel-9 notes

One actual formatter outputHey Sam, the beta looks good. I can join at 2:15, and I'll bring the Kestrel-9 notes.

Combine compatible rules, not competing ones

A useful bundle can pair one structure rule with exact-preservation and no-invention rules. Put the main output first, then the facts to preserve, then the additions to forbid. Avoid combinations such as “rewrite aggressively” and “preserve every word exactly.”

Testing boundary: these examples used synthetic English text in direct production-formatter calls, not end-to-end microphone tests. Separate synthetic-audio tests covered the transcription path; most structural requests did not pass every requirement. In these bounded tests, AI Smart Edit applied requested structure more often than Command Suite alone, but broad prompts still failed and model output can vary. Review every result against the source.

Use Command Suite for stronger transcription guidance, but do not expect Command Suite alone to perform structural transformations; the synthetic audio tests often preserved the words without applying the requested layout.

Step 7

Set the parts that affect everyday use

Work domain: choose the kind of work you do, such as Medical & Clinical, Legal & Compliance, Software Engineering, Sales, Customer Support, or General.

Typing Mode: turn this on when text does not land correctly in Citrix, RDP, VMware, terminals, EMRs, CRMs, or locked-down apps. DictaFlow will type character by character instead of relying on the clipboard.

Toggle mode: start and stop recording with one press instead of holding the trigger. Useful for longer dictation or accessibility.

Notes: DictaFlow notes stay local on the device. If you want synced notes, dictate into a notes app that already syncs, like Apple Notes, OneNote, Notion, or Google Keep.

Quick fixes

If something feels off

Mic not working?
Check Settings → Audio Input and your system microphone permission.
Shortcut conflict?
Change the dictation or AI Smart Edit trigger in Settings.
Remote app skips characters?
Increase Typing Delay under advanced settings.
Background noise?
Adjust Silence Sensitivity so quiet sounds do not turn into random text.

Ready to try it?

Download DictaFlow, open the app you already write in, and try one short sentence first.