Walk into the interviewwith the right questions
Most discovery interviews are prepared in the ten minutes before the call, with questions borrowed from a template that knows nothing about your product. Khint turns a research goal into a structured guide: open questions, follow-up probes, and a timebox, grounded in the epics and specs you already have.
Generic questions
get generic answers
“What are your biggest pain points?” is a question anyone could ask about any product, and it returns an answer you could have guessed. A good guide asks about the step users actually abandon, in the vocabulary they actually use, and it takes real preparation to write. Khint runs that preparation where your research goal is already typed: an agent on the selection, with your session context and your own indexed specs riding along.
The palette comes to the text, not the other way around
No window to open, no tab to switch, nothing to paste twice. Khint reads the selection where it is and writes the answer back in the same place.
The real palette: an agent runs on the selected research goal and the interview guide comes back in place.
From a research goal to a guide you can run
Preparing six discovery calls on why users abandon checkout.
Open a session for the study
Before anything else, start a Memory session and call it “Checkout discovery”. Every agent you run for this study is logged to it, and Khint keeps a compact running summary that rides along on each prompt. By the third interview the model already knows the hypothesis, the segment you are recruiting, and what the first two participants said. See how Memory works.
Write the goal in one paragraph, then select it
Type what you are trying to learn: the behaviour you are investigating, who you are talking to, the decision the research has to inform. Two or three sentences is enough. Select it where you wrote it, in your notes app, a Confluence draft, a Notion page, and press Cmd+Shift+K. Khint reads exactly what you highlighted and nothing else on your screen. See how agents work.
Run the interview guide agent
In Agents, save an agent named “Discovery interview guide”. No prompt yet? Type a description of what you want back and press Generate: an objective, a warm-up, open questions grouped by topic, a probe under each one, a closing question, and a timebox. Press Improve later to tighten it, for example to forbid leading and yes or no phrasing. The guide is pasted back where the goal was.
Ground the questions in what you already know
Index a tool in Knowledge and every run first searches your local index of Jira, Confluence, Notion and Linear, then hands the agent the closest excerpts. The questions come back referring to your actual flow and your actual wording rather than a generic funnel. The search costs no credits and the index stays on your machine. If the brief lives in a file, point the agent at the document instead of a selection.
Publish it where the team can read it
Connect Atlassian and the palette adds a Confluence Create page row, so the guide lands as a page your researcher, designer and engineer can comment on before the first call. One write per run, and the page URL comes back as the result, so publishing is always deliberate and always traceable. See how integrations work.
Goal to published guide, in one shortcut
Up to five steps in series: AI agents first, then the write that leaves your machine.
The second agent is the one worth keeping: it rereads the draft and rewrites anything that suggests its own answer, which is the failure mode a template never catches. Swap the steps for your own house rules, a consent and recording preamble, a translation step when the participants are not in your language, or a Notion page instead of Confluence. The space is pinned once in the workflow editor, so every later study is one keystroke.
What it does not pretend to do
A guide is preparation. The interview is still yours to run.
It does not sit in your call
Khint has no meeting bot and joins nothing. It prepares the guide beforehand and helps you work through your own notes afterwards, which is why nobody has to be told a recorder is present.
It reads what you point at
The selection you highlighted, the file you pinned, and the session summary you opted into. Not your screen, not your other windows, not your inbox.
Nothing is published by accident
An agent gives the guide back in place. A page is only created by an explicit write step or write row, and each run performs one write.
The index stays local
The Jira, Confluence, Notion and Linear pages Khint searches to ground your questions are indexed into a database on your own machine, never onto Khint servers.
Then the interviews happen
A guide is half the work. When the calls are done you have six sets of rough notes and a deadline, and that is the other half: synthesizing the interviews into cross-interview themes, then turning the themes into requirements and backlog items. It is the same session you opened in step one, so the summary that shaped your questions is the summary that reads your answers. Keep the guide agent and the synthesis agent in one pack and the whole study runs behind a single shortcut.
Common questions
What does the AI actually produce for a discovery interview?
A working guide rather than a list of themes: an objective line you can read out, a warm-up question, open questions grouped by topic, a follow-up probe under each one so you know what to ask when the answer is short, a closing question, and a rough timebox per section. You write the research goal in two or three sentences, select it, and the guide comes back where you typed the goal. Because you saved the agent, the next study starts from the same structure instead of a blank page.
How is this different from asking ChatGPT for interview questions?
A chatbot writes questions about the product it can infer from your paragraph. Khint searches your own material first: with a source indexed in Knowledge, the run looks through the local index of your Jira issues, Confluence pages, Notion pages and Linear issues, then gives the agent the closest excerpts. The questions come back naming your real flow, your real segment names and your real feature vocabulary. You also stay in the document you were writing in, so there is no paste out and paste back.
Can I stop it writing leading questions?
Yes, and it is worth making it a separate step. Put the rule in the agent prompt, then add a second agent to the workflow whose only job is to reread the draft and rewrite anything that suggests its own answer or can be answered with yes or no. Workflows run up to five steps in series and each step receives the previous step's output, so the checked version is what reaches the publish step.
Where does my research goal and product context go?
An agent sees the text you selected, plus the compacted summary of your active Memory session if you opted in, plus the few Knowledge excerpts that match once you have indexed a source. Memory sessions live in a local SQLite database on your Mac and are never synced to Khint's servers, and the workspace index is built on your machine too. Nothing about an unreleased feature leaves your device silently.
Do I have to publish the guide to Confluence?
No. The agent alone gives you the guide in place, and that is a complete flow. The publish step is there because a guide the team can comment on before the first call is a better guide: connect Atlassian or Notion and the palette adds a create-page row, or put that write at the end of a workflow so drafting and publishing happen on one shortcut. Every write is one write per run, and the page URL comes back as the result.
Prepare your next discovery round in minutes
Free with 300 credits a month, about 10 AI actions a day. No credit card. Save the guide agent once, reuse it on every study.