§ Context-aware
Context-aware dictation
Enhanced mode attaches three optional signals to every run: the frontmost app and its bundle ID, the active selection (via Accessibility), and a single window screenshot (only when you turn it on for that app class). Bounded payload, never silent.
Who this is for
Anyone who notices that the same sentence dictated into Slack and into a code editor should not be polished the same way. Especially developers, writers in multiple voices, and anyone with sensitive apps they want to exclude.
The three signals
Frontmost app
App name + bundle ID. Cheapest signal. Drives per-app polish profiles.
Selected text
Active selection read through Accessibility. Used so Enhanced understands what you are referring to. Requires Accessibility.
Window screenshot
Single-frame screenshot of the focused window. Opt-in per app class. Requires Screen & System Audio Recording.
App-class context allowlist
Settings › AI Processing › App context
- Enable app context
- Master switch for all three signals.
- Code apps
- VS Code, Cursor, Zed, JetBrains, Terminal, iTerm — context capture on by default when enabled.
- Mail apps
- Mail, Spark, Superhuman — separate toggle.
- Browsers
- Safari, Chrome, Arc, Firefox — separate toggle.
- Other apps
- Everything else. Disable to keep context off in apps you have not classified.
Bounded payloads
Dettivo enforces a per-context token budget and a payload size metric. Long selections are truncated. Screenshots are downsampled. The token budget is visible in Settings so you know exactly what gets sent if you have a hosted Enhanced provider configured.
Launching April 2026
Be first on the launch list.
Dettivo opens for new purchases in April 2026. One-time license, three Macs, lifetime updates.