A workspace decides how your dictation is rewritten before it is typed. Without one, a Slack message and a customer email come out in the same voice. With one, each gets treated appropriately.
Four categories of entries
The most common misunderstanding is that a workspace has four text boxes. It does not. It holds a list of entries, each with a title and a body, filed under one of four categories — and you can have as many of each as you like, switching individual entries on and off:
| Category | Answers |
|---|---|
| Tone adjustment | What voice should this be in? |
| Output style | How should it be shaped — prose, bullets, an email body? |
| Domain context | What subject matter is this, so terms are heard correctly? |
| Custom instructions | Anything else — your do's and don'ts |
Entries can also be files rather than typed text. Plain text, Markdown, CSV, JSON, XML, YAML, logs and PDFs are accepted, up to 10 MB, with PDFs read up to the first 50 pages.
How they reach the model
At dictation time, active entries are grouped by category and wrapped in labelled blocks that the system prompt knows how to interpret:
what the model receives[TONE ADJUSTMENT] …your tone entries… [/TONE ADJUSTMENT] [OUTPUT STYLE] …your style entries… [/OUTPUT STYLE] [DOMAIN CONTEXT] … [/DOMAIN CONTEXT] [CUSTOM INSTRUCTIONS] … [/CUSTOM INSTRUCTIONS]
Your raw transcript arrives separately, explicitly marked as speech to clean up rather than as instructions to follow. That separation is what stops a dictated sentence like "ignore the previous instruction" from hijacking the rewrite.
The settings people miss
Beyond entries, each workspace carries:
- Tags — what makes it activate automatically. See Auto-detecting workspaces.
- Vocabulary — terms to bias transcription toward. See Vocabulary and dictionary.
- Preferred language — force a spoken language instead of auto-detecting
- Preferred speech and text models — override your global choice for this context only
- Voice Only mode — skip AI refinement entirely and inject the raw transcript
- Default — the fallback when nothing else matches
cd dot dot into prose you did not want.What you start with
Four workspaces are created on first launch:
| Workspace | Set up as |
|---|---|
| Speech / Voice | The default, with AI refinement off — raw transcription |
| Chat | Casual tone, plain message, conversational |
| Professional tone, email body, polished | |
| Note Taking | Neutral tone, bullet list, concise |
How many you can have
The free plan allows five workspaces. Since four are seeded, that is the defaults plus one of your own. Pro and perpetual licences are unlimited.