They presented it in slides.You have to build from it
Forty slides, three of them appendices, half the scope living in bullet fragments written for someone who was in the room. Point a Khint agent at the file itself and every slide comes back as written requirements, in slide order, with the questions the deck never answered pulled out as their own list.
A deck is a talking format,
not a working one
Slides are written to be narrated. The bullet says “simplified onboarding” because the person presenting explained the rest out loud, the numbers are on slide 31, and the one constraint that changes the estimate is in a footer. Nobody can build from that, so the first days of the project go into retyping it: read a slide, work out what it actually asks for, write it down, remember which slides you already did. Khint makes the file itself the input of an agent, so the retyping is the part that stops. See how document input works.
A real workflow run: the steps execute one after another and the finished write-up lands in the tool the rest of the team already reads.
From their slides to your requirements
Turning a stakeholder deck for a customer portal into something the team can size.
Set the agent's input to Document, not selected text
In Agents, create the action that turns a deck into your team's requirement format, and set its input source to Document. That is the whole difference: the input here is a file you were sent, not something you can highlight on screen. No prompt yet? Type a one-line description and press Generate, then Improve it once you have seen a real result.
Point it at the deck, or pin the deck to it
A deck that arrives once per project is an Ask me each run agent: press Cmd+Shift+K and the palette opens a picker with your recent files and a browse button before anything is sent. The reference deck you will reread for six weeks is a pinned one: choose it once in the editor and the shortcut goes straight to work.
Let it read the slides, not a screenshot of them
The .pptx is opened on your machine and the text of every slide is pulled out in slide order, including the text inside shapes and tables. No pasting one screenful at a time into a chat window, no slides quietly skipped because you lost count at 23. A long deck is condensed once into a working summary that is cached against the exact content of the file, so rereading it later costs nothing new until the file changes.
Ask for the gaps, not just a tidy summary
A summary of a deck is still a deck. The line in the prompt that earns its keep is the one asking for what is missing: no acceptance rule, no volume, no error case, no owner. You get requirements you can size and, beside them, the short list you send back before anyone estimates anything. Pair it with the acceptance criteria generator and with non-functional requirements, which is where a deck is almost always silent.
Publish it where they can answer you
Connect Atlassian and the palette adds a Confluence Create page row: the write-up lands as one page the stakeholder can comment on, next to the open questions, instead of sitting in your notes app. One write per run, so publishing is always deliberate. See how integrations work.
One shortcut, deck to review page
A workflow can start from a document too, so the file is the thing the whole chain runs on.
Each step's result becomes the next step's input: the first agent turns the slides into requirements, the second reads that draft and marks what the deck left undecided, and the Confluence step publishes what comes out. Set the workflow's Starts from to Document and it will either ask you for the file each run or always use the one you pinned. Swap the last step for Notion or Slack depending on where your stakeholders actually read things. See how workflows work.
Two ways to get through forty slides
The reading is the same work. What differs is who does the mechanical half.
Slide by slide, by hand
- You copy a few bullets at a time into a chat window and lose the order they were in.
- The format drifts as the afternoon goes on, so slide 8 and slide 34 come out looking like different projects.
- Version 2 of the deck arrives and you cannot tell which parts you had already done.
- The gaps stay invisible, because a bullet that says nothing looks exactly like a bullet that says something.
The deck as the input
- Every slide is read in order, in one pass, from the file itself.
- The format is written into the agent once and holds for the whole deck.
- The prompt asks for open questions as their own section, so silence gets a name.
- The next deck is the same shortcut, with nothing to set up again.
What it will not do for you
Worth knowing before the deck lands, not after.
A picture on a slide stays a picture
What reaches the agent is text: the words in the boxes, the shapes and the tables. A screenshot pasted onto a slide, or an architecture drawing exported as one image, carries nothing to read. That slide belongs to the capture path, which reads pixels and costs credits from your monthly allowance.
Ask for the .pptx, not the .ppt
The modern PowerPoint format is read; the legacy binary .ppt is deliberately not parsed. Each failure comes back as itself, encrypted, too large, no text inside, unsupported, so you know whether to ask for a re-export or to split the file instead of staring at one vague error.
A long deck is condensed, not transcribed
Past roughly 10 KB of extracted text the deck is turned into a working summary once and reused from cache on later runs. That is what makes a big file cheap to come back to, and it is a real limit: when you need a requirement written from one specific slide with nothing paraphrased, run it on that section alone.
Nothing is filed behind your back
An integration write creates one page or one issue per run, after you confirm it. Forty requirements do not become forty tickets on their own, because the review is the part with the value: you read the set, cut what the stakeholder still has to answer, and file the rest.
Why not paste the slides into a chat window?
You can, and you will do it a screenful at a time, forty times, losing the slide order and rewriting the instructions on every batch. The file input removes the round trips: the deck is read locally on your machine and only the text your action sends reaches the model, the agent is saved once and reused on the next client's deck, and a Memory session started for the project carries the vocabulary into everything else you write that week, the follow-up mail, the epic description, the kickoff page. The mechanics of file input are covered in running AI actions on documents, and if you need the movement in the other direction, turning your own notes into slides, that is the PowerPoint diagram generator.
Common questions
Which presentation files can a Khint agent read?
Modern PowerPoint decks (.pptx) are read slide by slide, in slide order. The same document input also accepts Word, Excel and OpenDocument files, PDFs and plain text, so the agent you write for decks is not a one-format tool. The extraction happens on your machine with no model call, and if a file arrives with a wrong or missing extension, which happens constantly with exports from shared drives, the content is inspected rather than trusted so it still opens. The legacy binary .ppt format is the exception: it is deliberately not parsed, so ask for a re-export.
What does the agent actually see of a deck?
The text, in the order of the slides: titles, bullets, the words inside shapes, and table cells. What it does not see is anything that is only a picture, an image pasted onto a slide or a diagram exported as one flat graphic, and it does not see styling, so a deck whose meaning lives in its colour coding needs that meaning written somewhere as words. For a slide that is genuinely a picture, frame it with screen capture instead, which reads pixels.
What happens with a sixty-slide deck?
Above roughly 10 KB of extracted text the deck is condensed once into a working summary, cached against the exact content of the file and reused on every later run. Change the file and it is rebuilt on the next run; change nothing and there is nothing to rebuild, which is what makes a big deck cheap to keep coming back to. The honest caveat is that a condensed version is a summary rather than a transcript, so when you need every line of one section preserved, point the agent at that section on its own.
Can it create the Jira tickets straight from the deck?
Not in one sweep, and the page would be lying if it said so. An integration write creates one issue per run, after you confirm the draft on screen. What you get from the agent is the whole set of requirements as text, in the order of the deck: you read it, decide what is real, then file them one at a time from the palette, which drafts the issue and hands the key back to you. Turning a deck nobody has reviewed into forty tickets is how a backlog gets polluted in an afternoon.
Does the deck leave my machine?
The file is opened locally to read and is never uploaded anywhere. What reaches the model is the text your action sends, exactly as it would be for a paragraph you selected, and the condensed summary of a long deck is cached on your device. If you run the action inside a Memory session, that session lives in a local database on your machine and is not synced to Khint's servers, so an unreleased client scope is not sitting in a chat history somewhere.
Let the deck be the input
Free with 300 credits a month, about 10 AI actions a day. No credit card. Write the agent once and the next stakeholder deck is an afternoon instead of a week.