One AI toolkit forUX researchers
Interview notes into structured debriefs. Transcripts and screenshots processed where they are. One local memory per study, findings filed to Notion, and participant identifiers redacted before the model ever reads a word.
Eight interviews, forty pages of notes,
and the readout is due Friday
A study produces text faster than anyone can process it: session notes, exported transcripts, screenshots of the moment a participant got lost, and a repo page that was supposed to be updated after every round. The synthesis happens late, from memory, under deadline. And the obvious shortcut, pasting raw participant data into a browser chatbot, is exactly the thing your consent forms and your legal team say you can't do.
Khint keeps the whole loop behind one shortcut, inside the tools you already use. Select notes anywhere and run a saved AI Action on them in place, or point an Action at a transcript file directly. One Memory session per study absorbs every debrief and capture, so the synthesis is grounded in the whole study, and a per-Action switch redacts identifiers before anything reaches the model.
One study,
kickoff to readout
Every stage below is a shipped feature, not a roadmap slide.
- PlanRough questions → a usable discussion guideA study starts as a messy list of things you want to learn. Select it anywhere (a doc, an email, a Slack message) and run a guide-shaping Action in place: your saved prompt turns the list into ordered, neutral, open questions in your team's format. The prompt is yours, written once in the Action editor, or drafted for you from a one-line description with Generate.
- SessionOne Memory session per studyBefore the first interview, start a Memory session named after the study and make it active. From now on every debrief you run, every transcript you add, and every capture you take through Khint lands in that session and is compacted into one study context, kept in a local database on your machine. Interviews from different studies never blur together, because only the active session's context is injected into the Actions you opted in.
- DebriefRaw interview notes → structured debriefRight after the call, select your raw notes and run a debrief Action: what the participant did, what they said verbatim that matters, what surprised you, what to probe next time. Because the study session rides along, the third debrief already knows what the first two participants struggled with, so recurring observations get flagged as recurring instead of rediscovered.
- TranscriptsRun actions on the transcript file itselfGot an exported transcript instead of notes? An Action can take a document as its input: run it from the palette, pick the file (recent files are one click away), and the action works from the transcript without you selecting anything. Short files are read whole; long transcripts are condensed into a faithful digest first, so a 90-minute session doesn't break the run. You can also drop the file into the study session on the Memory page so the study context absorbs it.
- ScreensQuestion the screenshots, don't retype themUsability findings often live in pixels: a recording still, a participant's screenshot, a confusing screen you want to describe precisely. Extract text draws a region and puts the on-screen text on your clipboard. Capture & ask goes further: draw a region and ask questions about the screenshot in a chat, like which step of the flow the participant was on or what the error actually says.
- SynthesizeAcross interviews, not one at a timeWhen the sessions are done, run your synthesis Action with the study session's context on: it works from the compacted record of every debrief and transcript you logged, so themes are grounded in the whole study. Copy recap turns the session (summary plus every entry) into paste-ready markdown when you want the raw material in front of you. The full walkthrough of this step lives on the user interview synthesis page.
- ShareFindings land in the repo, not in a drawerFrom the same palette, Create page files the synthesis into your Notion research repository, and Append adds late findings to an existing page instead of spawning page four of the same study. A Slack post shares the highlights where the product team actually reads. Every write is something you trigger yourself, one at a time, and each one is logged back into the study session.
Participant data needs boundaries
a chat tab doesn't have
You promised participants their words would be handled with care, so the tool has to make that promise structural. An Action only ever sees the text you selected (or the one file you pointed it at), plus the active study session's compacted summary if you opted that Action in, and nothing else on your screen or disk. Study sessions live in a local database on your machine and are never synced to a server; what your team gets is what you choose to file to Notion or paste from Copy recap. There is no recording and no automatic transcription: nothing is collected unless you select it or capture it yourself. And for actions that routinely touch participant text, a per-Action switch replaces email addresses, phone numbers, SSNs, and IBANs with placeholders before the model ever sees the text.
The four pieces of the toolkit
Research workflows, one page each
The parts of the loop, in depth: how each one works with real Khint features.
Common questions from research teams
My notes are full of participant data. What does the AI actually see?
Only what you explicitly act on: the text you selected, plus (only if you tick the option on that Action) the compacted summary of the active study session. Khint does not read your other windows, your files, or your research repository. For actions that routinely touch participant text there is also a per-Action redaction switch that replaces email addresses, phone numbers, SSNs, and IBANs with placeholders before the model sees anything. It is pattern-based, a safety net rather than a compliance product, but it keeps the obvious identifiers out of the request. Study sessions themselves live in a local database on your machine and are never synced to a server.
Can I run an action on a whole interview transcript file?
Yes. An Action can be set to take a document as its input instead of selected text: run it from the palette and Khint asks which file, with your recent files one click away. Short files go to the model whole; a long transcript is first condensed into a faithful digest so the action works from a compact version of the file rather than failing on length. You can also drop transcript files straight into the active study session on the Memory page, so the study's context absorbs them.
Does Khint record or transcribe my interviews?
No. There is no recording, no meeting bot, and no automatic transcription. You bring the text: your own typed notes, an exported transcript file from whatever call tool you use, or a screenshot you deliberately capture. Capture only runs when you trigger it from the palette, and on macOS the screenshot is taken by the system's own screenshot UI. That is a deliberate boundary: nothing about a session with a participant is collected unless you select it or capture it yourself.
Can my research team share the same prompts?
Yes. Save your debrief, tagging, and synthesis prompts as Actions inside a Pack, then Share the pack with your team. Teammates who joined via your invite link see it as a read-only Team Pack and clone it into their own packs with Copy, after a preview modal shows every prompt in full. Everyone runs the same house-style debrief format behind the same shortcut, and each person can still tweak their own copy.
Is there a free plan?
Yes. The free tier includes 10 AI actions per day and 5 screen captures per day, no credit card. Paid plans start at €7/month: Pro at €29/month gives 100 AI actions a day, and every plan includes all the work integrations. Khint runs on macOS 13+ and Windows 10 22H2+.
Bring it to your next study
Free with 10 AI actions and 5 captures per day. No credit card; it runs on your own machine, next to the notes and the transcripts you already have open.