Context awareness
Context awareness lets Kalam look at the app you are typing into so AI polish produces text that fits — matching an email thread's tone, continuing a document, or respecting what's already in a field.
What it can read
All sources are configured per mode in Dictation → a mode → Read your surroundings.
| Source | What it captures | Limit |
|---|---|---|
| Active app content | Window title, plus the focused field's text, plus surrounding window content — depending on the depth you choose | App text ~2000 chars |
| Clipboard | Whatever plain text you copied before dictating | ~1000 chars |
| Selected text | Text you have highlighted when you press the hotkey | ~2000 chars |
| Date and time | The current local date and time | — |
App content has four depths: Off, Label (app name and window title only), Focused (plus the text in the field you are in), and Extended (plus surrounding window content). Password fields are never read.
How it is used
Context is captured only to inform the AI polish step — it never goes to the speech-to-text engine. When a dictation finishes, captured context is included in the single polish request so the model can match tone and content. If a mode captures context but has Improve off, Kalam still makes one AI call with just the cleanup baseline plus context.
Plan gate
Context awareness is a Pro feature (included in Trial and Max). On Free, the controls are locked and no context is captured.
Privacy
- Context is text, and it is sent only to the AI provider you configured (or Kalam-hosted AI on Max) as part of the polish request.
- If your AI runs locally, context stays on your device too.
- Sensitive apps block context completely. If the focused app matches your sensitive list, nothing is captured or sent for that dictation. See Sensitive apps and privacy.
Selection and app-content capture are most complete on Windows. Some Electron apps expose limited text, and elevated (run-as-admin) windows can block reading. macOS capture requires Accessibility permission.