Interactive simulation
How doki handles Cursor
Inbox0Waiting for the agent…
Which layout do you want for pricing?
Click an option or press 1, 2, or 3
- 1Finds the sessionReads the app’s local database, the CLI, and Cursor Cloud through the API.
- 2Tells youThe question rises to the Inbox with its options.
- 3You answerA key, your voice, or the iPhone. Two seconds to undo.
- 4Delivers the replydoki replies through the API · or cursor-agent -p --resume
- Cursor aloneYou are inside the editor, the agent belongs to Cursor, and you read the diff right there.
- Cursor plus dokiYou want that agent next to Claude or Codex, you want to answer without bringing Cursor forward, or you want to follow it from your phone.
| Cursor | doki | |
|---|---|---|
| Editor and in-IDE diffs | Yes | Does not replace the editor |
| Cursor app chats | Inside the editor | Shown from the local database |
| Cursor CLI | cursor-agent in the terminal | Replies with cursor-agent -p --resume |
| Cursor Cloud | Cursor API | The same API, with an optional worker on the Mac |
| API agent in your folders | You use the API or the app | You create it in doki: model, mode, and git |
| Typing into Cursor’s window | Keyboard in the editor | Desktop bridge or Accessibility |
Three ways to have an agent in Cursor
The desktop app stores its chats in a local database. doki reads them and puts them in the Inbox with the same status as any other session. The CLI is a separate process: doki detects it and resumes it when needed. Cursor Cloud lives on the API; doki talks to that API and can keep a worker on your Mac.
Three products under one name, and three places a question can sit. The Inbox is there so you do not have to remember which of the three is waiting.
Working without bringing Cursor forward
“On my Mac (API)” creates a Cursor agent that operates on your folders. You message it from doki. You choose the model, whether it plans or only edits, and what it should do with git when it finishes. Cursor is still the editor when you want to read the diff slowly. The “should I continue?” loop does not require an app switch.
If the session is in the window, doki can type there through the desktop bridge or Accessibility. That permission has to be granted on the Mac. Without it, the CLI and API paths still work.
Cursor is rarely the only agent
A common split is Cursor on the interface and Claude Code or Codex on the logic. Comparing them as a forever choice is the Claude Code vs Cursor page. The point here is different: even if you pick Cursor for a task, that agent’s question competes with the others.
When the agent finishes, doki can review the diff and return a verdict: ready, review, or don’t merge, with risks and detected secrets. That step is described in diff review.
Where the data stays
Cursor’s chats stay in Cursor’s database, on your Mac. doki does not upload them to its server. Optional provider keys, including Cursor’s if you add it, go to the Keychain, not into a file in the repo.
A doki license unlocks the app on up to three Macs and keeps working offline for days. Cursor model usage is billed where you already pay for it.
Frequently asked questions
Does doki replace the Cursor editor?
No. Code is still edited and reviewed in Cursor. doki steers the agents and tells you when they ask or finish.
Does it see the chats I already have in Cursor?
Yes. It reads the app’s local database, plus the CLI and Cursor Cloud if you use them.
Can I create a Cursor agent without opening the app?
Yes, through the API, on your folders. In doki you pick the model, the mode, and how git is handled.
Do I need a Cursor key?
Only for the paths that use the API. Local chats and the CLI are discovered on the Mac without that key.