AI Enhancement settings
Full reference for the AI Enhancement panel — models, prompts, providers, Screen Awareness.
Reference for the Settings → AI Enhancement panel. For the conceptual overview, see Features → AI Enhancement.
Layout
Top to bottom:
- Enhancement toggles — master on/off, plus Screen Awareness.
- Model selection — Recommended cards, then All models grouped by provider.
- Enhancement prompt — built-in prompts and your custom prompts.
The prompt section only appears when enhancement is enabled.
Settings → AI Enhancement panel, scrolled to show enable toggle, recommended model cards, and the start of the prompt list
Choosing a model
Recommended highlights fast, cheap models that work for most enhancement:
| Model | Provider | Notes |
|---|---|---|
| GPT-OSS 20B | Groq | Open model on Groq's fast inference |
| GPT-OSS 120B | Groq | Larger variant, higher quality |
| Gemini 2.0 Flash Lite | Fast, has free tier | |
| LFM2 24B A2B | OpenRouter | Cheap and fast |
Cards show Selected, Ready, or Setup Required. "Setup Required" → click to start the API-key flow.
Below recommended, All models lists everything. Type in the search field to filter by name. Provider groups auto-expand on match.
API keys
Built-in providers (OpenAI, Anthropic, Google, Groq, OpenRouter, Together) use one active key per provider for AI Enhancement.
To switch active keys: Settings → API Keys, select a key on that provider's row. Cloud Transcription has its own separate key chooser if you want different keys for transcription and enhancement.
Where to get keys: see API Key Setup.
Built-in prompts
| Prompt | What it does |
|---|---|
| Clean Transcript | Light cleanup, original shape preserved |
| Professional Rewrite | Polished business communication |
| Casual Message | Light cleanup, chat tone |
| Email Draft | Email body, no greeting / closing |
| Bullet Summary | Concise bullet points |
| De-Rambler | Stream-of-consciousness → coherent draft |
| Translate to Spanish | Clean + translate to neutral Spanish |
| Translate to Jamaican Patois | Clean + translate to readable Patois |
All share a cleanup baseline: fix obvious transcription errors, remove fillers when meaningless, preserve intent.
Click any prompt card to make it the active default.
Custom prompts
Click Add custom prompt:
- Name — what you'll see in the list.
- System — instructions for transforming the transcript.
Critical: write the system prompt as "transform this transcript into X," not "respond to this transcript." Otherwise the model answers questions in the transcript instead of cleaning them up.
See Features → AI Enhancement for examples.
Custom providers
For self-hosted endpoints, Azure OpenAI, LM Studio, Ollama, or any OpenAI-compatible API:
- Click Add custom provider.
- Set the base URL.
- Choose auth: Bearer, API key header, or none.
- Link a saved Custom key from Settings → API Keys if auth is needed.
- Save.
Models for the new provider appear in All models under its name.
Screen Awareness
Toggle for the optional text-only screen context feature. See Permissions for privacy details and Features → AI Enhancement for the overview.
Toggling on prompts for Screen Recording permission if Whiskers doesn't already have it.