Type an idea into the palette and Khint plans it out of your own agents and the tools you've connected. You read the plan before anything runs, and a run worth keeping becomes a shortcut.
Ran once · keep it and it becomes a shortcut
The same entry sits in Capture & ask: one button under an answer turns it into a plan built from your own agents and connected tools. The steps light up inside the thread as they run, and what they created lands on your clipboard.
The card names every step and the text it will work on, and resolving it costs nothing. The payoff is behind the window rather than inside it: the board gains the ticket, which is the only thing a chat window can't do for you.
A plan isn't a guess at what you might want. It's composed of two inventories you control, it reports whatever it couldn't reach, and it stays a one-off until you decide otherwise.
The planner reads two lists: the agents in your active pack, and the write actions of the tools you've connected. It composes those. It can't invent a capability Khint doesn't have, and it never reaches a tool you haven't signed into.
The search box filters what you saved. The idea bar describes what you have not. They stay separate on purpose, so a search is never mistaken for an instruction.
Anything the plan can't reach is listed as unmet rather than quietly dropped, so you find out on the card instead of three steps into a run that ends nowhere.
A planned run is a one-off: nothing is saved unless it worked and you say so. Then Khint writes the steps out as real agents and a real workflow with its own key, so the second time is one press. A kept plan also keeps the surface it was born on: planned from a capture, it re-runs from a region on screen.
Planning and running are two separate presses, and the card between them is the whole point. A plan you don't like costs you an Esc.
Security and privacyType it in the palette, say it with the palette open, or point at something already on your screen. All three end on the same plan card.
The palette opens on the idea bar, focused and empty. Describe the work and press ⏎.
With the palette open, your dictation shortcut fills the idea bar instead of typing into the app behind.
See voiceCapture & ask carries the same entry: act on an answer, or on a fresh capture before any chat.
See CaptureThe idea bar composes what these three give it, and hands back a shortcut you keep.
The palette opens on a "Tell Khint what to do" input. Type what you want done in plain language, for example "summarize this and file it in Jira", and press Enter. Khint turns the sentence into a plan: a one-off workflow built from your own agents and the tools you've connected. You see the plan before anything runs, confirm it, and it runs. A chat window would give you text to copy; this gives you the work done, in the tools your team already reads.
No, and that's the point. The idea bar is the way in for work you haven't set up yet: Khint writes the steps for this run. If the result was good it offers to save them, so the throwaway run becomes a reusable agent or workflow you can trigger by shortcut afterwards. You end up with a library you described rather than one you assembled.
The agents in your active pack, plus the write actions of the integrations you've connected: create a Jira issue, create a Confluence page, create a Linear issue, create a Notion page, post to Slack, draft a Gmail email. Anything it can't cover with those is reported on the plan card as unmet instead of being quietly dropped, so you never discover a missing step mid-run.
Both. With the palette open, your dictation shortcut sends the transcript to the idea bar instead of pasting it into the app behind: the text is filled in, never fired, so you read back what Khint heard and press Enter yourself. Capture & ask carries the same entry point from the other side, a "Do this" button on an answer or on a fresh capture, where Khint plans from what it read from the screen and puts the created issue key or page link on your clipboard.
The run does; the planning does not. Your selected text is never sent to the planner: it only ever sees the sentence you typed, your agent names, and which tools you've connected. The plan card then shows you the text the run will actually work on before you confirm.
Planning is free and never spends credits, so a plan you cancel costs nothing. Running it costs what the same work would cost as a normal run: one action's worth of your monthly credits per AI step, and nothing extra for the steps that write to your tools.
Free with 300 credits a month. No credit card. macOS 13+ or Windows 10 22H2+.
Create issue
Post the link
Confluence
Linear
Notion
Gmail