Confluence · Team Velocity Gadget
One screen is four or five slices
Ask the vision chat what a user can actually do here and the screen comes back as an inventory. Each line is a story-sized piece of work rather than one giant ticket nobody can estimate.
A screenshot in,user stories out
“Build something like this” comes with a screenshot, not a backlog. Khint captures the screen, lets you interrogate it with a vision chat, and turns it into well-formed user stories from one keyboard shortcut, without retyping a single feature.
One screen is never
one story
A list view, a filter, an empty state, a create flow: a real screen hides four or five slices. Capture it once, keep it pinned, and keep asking until the screen is inventoried, instead of squinting at a PNG and typing what you think you see.
The real Capture & ask window. The screenshot stays pinned, so a follow-up question costs no second upload.
One screen, start to a backlog
A “build something like this” screenshot, step by step.
Start a session for the thing you're scoping
Open Memory and start a session: “Notifications revamp”. Every capture and agent you run is logged and compacted, so as you mine screen after screen the model holds the whole feature in context and your stories stay consistent with each other. See how Memory works.
Capture the screen you were told to copy
A stakeholder drops a competitor's notification settings screen in Slack and says “we need something like this”. Open the palette, pick Capture & ask, drag a box around it. Emailed PNG or a phone photo of a whiteboard? Drop it into the Capture tab instead.
Break the screen into capabilities
Ask the vision chat in plain language: “List the distinct things a user can do on this screen, each as its own line.” The screenshot stays in context, so you keep going — what can they toggle per channel, what's the empty state — until the screen is inventoried into the slices that should each become a story.
Shape the inventory into user stories
Save a User stories agent once: rewrite each line as “As a [role], I want [goal], so that [benefit]” and add a note on what's ambiguous. Select the inventory, hit Cmd+Shift+K, run it. Don't want to write the prompt? Hit Generate and it drafts one from a one-line description.
File the stories where the team grooms them
With Jira or Linear connected, the palette grows write rows: a Create issue step drafts each story as a ticket and hands the key back. Or chain it — a workflow can inventory, write the stories, then file the issue in one run.
What a capture is worth
The screen broken into slices, a second question that costs nothing, the two rows that start it, and the workflow that ends in your tracker.
Ask again without paying twice
The screenshot stays pinned to the conversation, so a follow-up is another turn on the same image rather than a second capture.
Two rows, and one of them opens nothing
Extract text reads the region and closes. Capture and ask opens the chat instead, which is the one you want when a screen has to be interrogated.
Inventory, stories, filed
The same pass can be a workflow. The capture is the trigger, one agent inventories the screen, the next writes the stories, and the last files them.
Create issueStories are the start,
not the end
A drafted story is reusable text, so the next agent can read it. Chain the whole pass and one screenshot becomes filed tickets in a single run.
See workflowsA follow-up agent writes the acceptance criteria
Or chain it: up to five steps in one run
Step one inventories the screen
Step two writes the stories
Step three files the issue and hands the key back
The session keeps the screens, stories and keys
Claude Desktop and Codex read it over MCP
Local SQLite, never synced to Khint
The rest of Khint
A capture is one way in. These three are what reads it, shapes it and files it.
Common questions
How is this different from extracting requirements from a screen?
Same capture gesture, different artifact at the end. Requirements come out as system-shall statements describing what the software must do; user stories come out in the role-goal-benefit shape your team grooms and estimates: "As a [role], I want [goal], so that [benefit]". Khint doesn't hardcode either one: the shape is whatever your saved agent's prompt asks for. Write the prompt to emit a numbered list of user stories and that's what every screen gives you back.
Can it draft more than one story from a single screen?
Yes, that's the point of using the vision chat first. A real screen is usually several stories: a list view, a filter, an empty state, a create flow. Pick Capture & ask from the palette, drag a box around the screen, and ask the model to break it into the distinct capabilities a user gets from it. Then run your user-story agent on that inventory so each capability becomes its own story, instead of one giant ticket nobody can estimate.
What if the screenshot was shared in Slack or emailed to me?
Drop the file into the Capture tab. PNG, JPG, WebP, HEIC, GIF, BMP, and TIFF go through the same flow as a live region capture, so a competitor screen a stakeholder pasted into Slack, a designer's exported mockup, or a phone photo of a whiteboard all work the same way. You get the same text out, ready for your story agent.
Where does the screenshot and my text go?
Captures go to the vision model so it can read the image; agents send only the text you selected, plus, if a Memory session is active, the compacted session summary. Memory sessions live in a local SQLite database on your Mac or PC and are never synced to Khint's servers. Khint's free tier covers 300 credits a month, about 10 AI actions a day with no credit card; paid plans start at €7/mo, with Pro at €29/mo for 4,000 credits a month; every plan includes all the work integrations.
Try it on the next screenshot you're handed
Free with 300 credits a month, about 10 AI actions a day. No credit card. Capture the screen, ask the vision chat, keep the stories.