Four vendor demos,one honest scorecard
A tool selection is decided by evidence you collect in fragments: a pricing grid on screen for eight seconds, a limit mentioned in passing, notes typed half-blind during a demo. Khint captures each vendor's claims as you see them, keeps one work session per demo, and scores every vendor against the same saved criteria, so the recommendation you publish traces back to what was actually shown.
The last demo always
sounds like the winner
Vendor evaluations drift because the evidence is uneven: the first demo lives in three-week-old notes, the last one in vivid memory, and the crucial caveat (the row cap, the tier that actually includes SSO) was on a slide nobody kept. By the time you assemble the comparison, you are reconstructing, and reconstruction favors whoever presented best. Khint's three pillars, Agents, Capture, and Memory, attack exactly that: claims are captured the moment they are on screen, each vendor's session is a separate record, and one saved scoring prompt makes the write-ups comparable instead of impressionistic.
From first demo to signed-off decision
One selection process, four vendors, no evening of reconstruction.
Write the criteria down once, as an Action
Before the first demo, turn your evaluation criteria into a saved Action: “Score these demo notes against security, data migration, reporting, support, and total cost: one line per criterion, met / partial / gap, with evidence.” Type that description and press Generateif you don't want to write the prompt yourself. This one prompt is what will make four different demos comparable.
Start a session per vendor
As each demo starts, start a Memory session named for the vendor. Every capture and every AI call you make during the demo is logged to it, and its compact summary rides along on later calls, so when you draft the final recommendation, the model is grounded in what this vendor actually showed, not in your fading recollection of it.
Capture what the salesperson moves past quickly
The pricing grid shown for eight seconds, the plan-limits table, the admin screen: hit Cmd+Shift+K and run Extract text: draw a rectangle and the text is on your clipboard and in your session. Something unclear on their dashboard? Capture & ask opens a chat about that region of your screen while it is still up. How Capture works →
Score the demo while it is fresh
Right after the call, select your raw jottings (abbreviations, half-sentences, all of it) and run the scorecard Action. Back comes a structured line per criterion, drawn from the selected notes plus the captures already in the session, pasted where you were typing. Ten minutes per vendor instead of an evening of reconstruction at the end.
Draft the recommendation and publish it
With all demos scored, select the scorecards and run a second saved Action that drafts the recommendation: ranked options, key risks, the call you are making. Edit it in place, then run the Confluence Create page row and choose use the text as-is: the decision lands in your space for the steering committee. How publishing to Confluence works →
Demo jottings in, comparable scorecard out
The criteria come from the prompt you saved: this is one example of what a scorecard Action might return.
Because every vendor's notes run through the same saved Action, “partial” means the same thing on all four scorecards, and each line carries the evidence, so a stakeholder who challenges the migration score can be pointed at the captured limits table it came from. This is the buying side of two jobs Khint also covers from the other direction: analyzing competitors is about products you compete with, and drafting RFP responses is what the vendors on your shortlist are doing right now.
Why not just ask a chatbot to compare them?
A chatbot compares vendors from its training data and their marketing pages, not from what the sales engineer actually showed you, under your contract terms, against your criteria. Khint works from your evidence: the captured pricing grid, the limits you extracted, the notes you took, all logged per vendor in Memory and carried into each AI call as compact context. The scoring prompt is written once and runs identically on every demo, and the result pastes back into the document you are already writing: no re-pasting context into a chat window between calls, and no vendor demo accumulating in a chat history. When the decision is made, the Confluence integration publishes it where the steering committee already looks.
Common questions
How does Khint keep four vendor demos from blurring together?
With one Memory session per vendor. Start a session before each demo (“CRM eval: Vendor A”) and everything you do during it lands there: captured screens, jotted notes run through your scorecard Action, questions you asked about their dashboard. Sessions are separate, so Vendor A's caveats never bleed into Vendor B's scorecard, and you switch the active session in one click before the next call. When it is time to compare, each session is a complete, honest record of what that vendor actually showed you.
Can Khint read a vendor's pricing table or limits page?
Yes. The palette's Extract text row lets you draw a rectangle over any region of your screen (a pricing grid, a plan-limits table, a slide in their deck) and the text lands on your clipboard and in your active session. If you want to interrogate rather than copy, Capture & ask opens a chat about the captured region, so you can ask “which plan includes SSO?” while the pricing page is still on screen.
Is the scorecard format built into Khint?
No: it is an Action you write once, which is the point. Save a prompt like “Score these demo notes against our criteria: security, data migration, reporting, support, total cost. One line per criterion: met, partial, or gap, with the evidence.” If you don't want to write it from scratch, type that description in the Action editor and press Generate, then refine it with Improve. Because the same saved prompt runs on every vendor's notes, the scorecards come out comparable instead of shaped by whoever's demo was most recent.
Vendor calls are under NDA. Where do my notes go?
Memory sessions live in a local SQLite database on your machine and are never synced to Khint's servers. When you run an Action, the model sees the text you selected plus the compacted summary of your active session if that Action opted in, nothing else on your screen or disk. A capture sends that one screenshot for text extraction, and the result is stored locally. If notes carry contact details, the per-Action redact-PII switch replaces emails and phone numbers with placeholders before the model sees them.
Can I turn the final comparison into a slide for the steering committee?
On a Mac, yes. Enable the PowerPoint companion on the MCP page, paste your comparison summary onto a slide, select it, and run Transform to diagram from the palette. Khint lays the options and criteria out as a clean comparison table (there is a dedicated comparison framework that frames the recommended option) and you save it back as native, editable PowerPoint shapes, not a screenshot.
Decide on evidence, not on the best demo
Free with 10 AI actions and 5 captures per day. No credit card. Save the scorecard Action before the first vendor call.