AI polish
Improve with AI is an optional step that rewrites the raw transcript before it is typed. It removes filler, fixes punctuation and grammar, and applies your writing instructions — so spoken rambling becomes clean text.
When it runs
Polish runs after transcription, only when the active mode has Improve with AI on (or captures surroundings) and an AI model is available. One dictation makes at most one polish call. If the model fails or refuses, Kalam falls back to the raw transcript.
Per-mode controls
In Dictation → a mode → Improve the text:
- Improve with AI — the master toggle for that mode.
- AI model — which language model to use (defaults to the Base mode's choice).
- Writing instructions — how you want the text written: tone, structure, formatting, sign-offs, and so on.
Leave instructions empty and Kalam applies a sensible cleanup baseline: remove fillers, fix punctuation and capitalization, preserve meaning and length, invent nothing.
Which AI models can polish
| Source | Models | Plan |
|---|---|---|
| Your own key (BYOK) | Groq, OpenAI, Anthropic, Gemini, OpenRouter, or a custom OpenAI-compatible endpoint | Pro / Trial |
| Kalam-hosted | Kalam picks a hosted model for you | Max |
Keys and custom endpoints live in Settings → AI & Models. On Max, no key is needed — polish runs on Kalam's hosted AI within your monthly quota.
What the AI sees
When polish runs, Kalam sends the language model:
- Your writing instructions (or the cleanup baseline).
- The transcript.
- Any surroundings context the mode captures — app content, clipboard, selection, date/time. See Context awareness.
It does not send your audio, your history, or your dictionary. Audio is only ever sent to a speech engine, and only if you picked a cloud one.
If the focused app is on your sensitive list, polish and context are skipped for that dictation entirely.
Voice editing
A separate voice editing hotkey lets you highlight text and speak an instruction ("make this more formal") to rewrite the selection with the active mode's AI model. Pro and above, Windows. See Hotkeys and voice.