What are workspaces?
A workspace is a profile that teaches InkSpoke how to write for one context. It bundles the words you use, the tone you want, and the background knowledge the AI should keep in mind — then applies all of it automatically based on the app or window you're dictating into. Legal dictation reads nothing like a Slack message, and workspaces are how InkSpoke tells the difference.
The problem workspaces solve
Without any context, InkSpoke refines every utterance the same way. That's fine for a quick note, but the moment your work has its own vocabulary and conventions, a one-size-fits-all polish starts to grate:
- Dictating a court filing, you want formal prose, correct legal terms, and no casual contractions.
- Dropping a line in Slack, you want it loose and short — contractions welcome, emoji fine.
- Writing code comments, you want terse, technical phrasing and your API names spelled right.
A workspace captures those expectations once. After that, when you dictate into your legal editor InkSpoke reaches for the legal workspace, and when you switch to Slack it switches too — no menus, no re-configuring per sentence.
What's inside a workspace
Every workspace is built from three ingredients:
┌────────────────────────────────────────────────────────────┐
│ Workspace: "Legal — Contract Drafting" │
├────────────────────────────────────────────────────────────┤
│ 🏷 Tags contracts, wordperfect, clio, legal │
│ → auto-match the right app/window │
│ │
│ 📖 Vocabulary indemnitor, force majeure, arm's-length │
│ → recognized by speech + known to AI │
│ │
│ 📚 Knowledge Tone Adjustment · Output Style · │
│ entries Domain Context · Custom Instructions │
│ → shape how the AI refines your words │
└────────────────────────────────────────────────────────────┘
Tags — how a workspace finds the right moment
Tags are labels like slack, vscode, or contracts.docx. When you dictate, InkSpoke scores
each active workspace's tags against the window you're typing into and picks the best match, so the
right workspace loads on its own. You'll see which one resolved on the
listening overlay. The full scoring rules — and how
to override the choice — live in
Smart matching & precedence.
Vocabulary — teaching InkSpoke your words
Vocabulary is a list of the terms, names, and jargon specific to this context. It does double duty:
- It's fed to the speech model as a hint, so unusual words are transcribed correctly instead of turned into the nearest common word.
- It's passed to the AI as known terms, so refinement doesn't "correct" your jargon into something wrong.
Knowledge entries — four kinds of guidance
Knowledge entries are where you tell the AI how to write. They fall into four categories, and each one steers a different part of the refinement:
| Category | What it steers | Example |
|---|---|---|
| Tone Adjustment | The voice and register of the output | "Formal, no contractions, third person." |
| Output Style | Formatting and structure | "Prefer numbered lists; keep sentences under 20 words." |
| Domain Context | Background the AI should assume | "This firm represents commercial landlords in disputes." |
| Custom Instructions | Any other rules for this context | "Always spell out statute names in full on first mention." |
You can build an entry a few ways: type it in, paste from your clipboard, start from a built-in
preset, or extract it from a file. File extraction reads plain-text and structured formats
(.txt, .md, .markdown, .csv, .json, .xml, .yaml, .yml, .log) as well as .pdf,
pulling the text out so a reference document can become instant context.
Each entry can be toggled active or inactive, so you can keep a library of context and switch pieces on and off without deleting them.
A workspace can also pin its own language, speech model, and text (refinement) model, so one context can run a different setup than the rest of your app. Those knobs — and how to tune every field — are covered in Create & tune a workspace.
How workspace content reaches the AI
When AI Refinement is on and a workspace has resolved, InkSpoke assembles that workspace's active knowledge entries into the refinement prompt, grouped under labeled sections:
[TONE ADJUSTMENT]
Formal, no contractions, third person.
[OUTPUT STYLE]
Prefer numbered lists; keep sentences under 20 words.
[DOMAIN CONTEXT]
This firm represents commercial landlords in disputes.
[CUSTOM INSTRUCTIONS]
Always spell out statute names in full on first mention.
That block rides along with your transcription so the model refines your words with your context in mind:
The content budget
Assembled content is capped so a huge reference file can't blow past the model's limits. The cap is
governed by MaxContentCharacters (default 40,000 characters), set on the Workspaces page.
If your entries exceed it, InkSpoke truncates at the last complete sentence before the limit and
appends a note like [... truncated, N chars omitted], so you never get a sentence cut mid-word.
Knowledge entries shape the AI refinement step, so they only apply when refinement runs. If AI Refinement is turned off — globally, or for a specific workspace — InkSpoke injects your raw transcription verbatim and no workspace prompt data is used. See How refinement works.
Which workspace applies?
For any given dictation, InkSpoke resolves exactly one workspace, in a fixed order of priority:
- Pinned — a workspace you've pinned always wins.
- Overlay override — a workspace you pick from the listening overlay, just for this session.
- Smart match — the best tag match for the current window (when smart matching is on).
- Default — the workspace you've flagged as default, as a fallback.
This is only a preview — Smart matching & precedence walks through each level and the scoring behind step 3.
The global layer
Some things you want in every workspace — your name and role, your company, a handful of terms you use everywhere. Rather than copy those into each workspace, InkSpoke has a global layer: your Personal Context and a global vocabulary that can contribute to every dictation.
Each workspace decides how it relates to the global layer — it can inherit it, force it on, or force it off — for both personal context and vocabulary independently. Vocabulary from the global layer and the workspace is merged, de-duplicated, and capped at 50 terms. This lets a workspace either build on your shared setup or stand completely on its own.
You'll find the details, including how to write good personal context, in Personal context & dictionaries.
Workspaces behave identically on Windows, macOS, and Linux. Smart matching relies on the operating system reporting the active window and app, which InkSpoke reads on each platform, but the tags, vocabulary, entries, and precedence rules are the same everywhere.
Next steps
- Create & tune a workspace — build one with the guided wizard, then adjust every field.
- Smart matching & precedence — how InkSpoke picks the right workspace, and how to override it.
- Personal context & dictionaries — the global layer, personal context, and word substitutions.
- How refinement works — what happens to your words after they're transcribed.