AI chat ⌃ S
An AI chat overlay that reads your screen.
Hit ⌃ S anywhere on your Mac. A floating panel appears, voice or typed, optionally screen-aware. On-device, free.
Three keystrokes, no setup.
- 01
Summon with ⌃ S.
A floating panel appears over the active window. The same shortcut closes it. It stays out of your way until you ask for it.
- 02
Ask by voice or type.
Quick-ask mode pastes the answer at your cursor and dismisses. Multi-turn mode keeps the panel open with full conversation history. Both run on Apple Foundation Models on-device.
- 03
Toggle screen context for visual queries.
Vision OCR reads your active window, locally on the Neural Engine. Ask about a contract, a Figma frame, a paper, a pricing page, an error in your IDE. Off by default; you opt in.
Built for the way specific people work.
- EngineersAsk about the file under your cursor, generate test cases, rewrite a docstring, explain a regex. Reads your IDE locally.See how →
- DesignersSpot weak hierarchy, suggest tighter copy, describe a design's voice, all from what's on screen in Figma.See how →
- StudentsOpen a 40-page PDF, ask for the key argument or a one-paragraph summary. The reading doesn't leave your Mac.See how →
- FoundersReading a competitor pricing page or your own deck, ask for tier extraction or tighter slide copy.See how →
- ResearchersOpen a paper, ask for the methodology or limitations. The PDF doesn't leave your Mac.See how →
Versus the obvious alternatives.
| Feature | Dollop | Cluely | ChatGPT desktop | Claude desktop | Apple Writing Tools |
|---|---|---|---|---|---|
| On-device (no upload) | ● | ○ | ○ | ○ | ◐ |
| Screen-aware (Vision OCR) | ● | ● | ◐ | ◐ | ○ |
| Voice input | ● | ◐ | ● | ○ | ○ |
| Multi-turn conversation | ● | ● | ● | ● | ○ |
| System-wide overlay (any app) | ● | ● | ◐ | ◐ | ◐ |
| Free | ● | ○ | ◐ | ◐ | ● |
| Resident RAM | 0 GB | cloud | ~150 MB | ~150 MB | 0 GB |
● Yes · ◐ Partial · ○ No. Verified May 5, 2026.
Questions, answered.
Is screen capture safe? What gets seen? +
Screen context is off by default. You toggle it on per-conversation. When on, Vision OCR reads the front window only, processes it on Apple's Neural Engine locally, and discards the image after the response. Nothing is uploaded. Cluely was flagged in early 2026 for screenshot capture that routed through their cloud; Dollop's posture is fundamentally different by architecture.
Does it work like Cluely? +
The form factor (a floating overlay, screen-aware) is similar. The architecture is opposite: Cluely processes screenshots in the cloud and bills $20/month; Dollop runs entirely on-device on Apple Foundation Models and is free. Full comparison.
Can the chat keep history across sessions? +
Yes. Multi-turn chats persist in the local Dollop library, searchable across past conversations. You can delete any chat at any time; "delete" actually means delete (no soft-delete on a server you don't own).
How fast is it? +
First token in tens of milliseconds, full response in well under a second for typical queries. Comparable to raw Apple Foundation Models speed; no cloud round-trip latency.
Can I use it in the terminal or in code editors? +
Yes, system-wide. The floating overlay sits over Cursor, VS Code, Xcode, Terminal, iTerm2, anywhere. Ask about the file you're looking at and the screen-context layer reads it locally.
What's the difference between AI chat and AI Mode? +
Same model, two interaction modes. AI Mode (quick-ask) takes a voice question, returns the answer at your cursor, and dismisses. AI Chat (multi-turn) opens the floating overlay and stays open while you keep the conversation going. Both use ⌃ S as the entry hotkey.
Hardware requirements? +
Apple Silicon (M1+) on macOS 26 (Tahoe) with Apple Intelligence enabled.
All the convenience of an AI overlay. None of the cloud routing. Free.
Download for Mac