The AI inside Jira startsonce the ticket exists
Atlassian Intelligence is switched on for your site and works on Atlassian content: the issue, the page, the comment thread. A product owner's day starts one step earlier, with notes in a meeting app, a screenshot of a staging bug, a Slack message, a client call. Khint is the layer above all of that: select whatever you have on screen, press one shortcut, and a saved agent turns it into a real Jira issue or Confluence page.
The hard part happens
before Jira sees it
Summarising a long comment thread is genuinely useful, and the AI built into Atlassian does it where the thread already is. But nothing you refined this morning is an issue yet. It is a block of notes, a screenshot someone pasted, a paragraph of a client mail. The pass that turns that into a well-formed ticket is the one that eats the afternoon, and it happens outside Atlassian. An agent is a prompt you saved once, it runs on whatever text you have selected, and the Atlassian integration files the result without a browser tab.
Describe it in the idea bar
Type what you want in plain language. Khint builds a plan out of your own agents and your connected tools, shows it to you, and runs nothing until you confirm.
Say it out loud
Tap your voice key, speak, tap again. The text lands at your cursor in any app, cleaned up in the same pass. An agent or a workflow can listen on its own key.
Point at your screen
Draw a region over anything on screen, even an image or a video call. Khint reads the text into your clipboard, or opens a chat you can keep asking questions in.
The ways into Khint: a selection in any app, a region of the screen, a document, or your voice. None of them is an Atlassian object yet, which is the point.
A refinement session, filed before the next call
Tuesday, 40 minutes of refinement, three things to get into Jira and Confluence afterwards.
Open a session for the epic
Before the call, start a Memorysession and name it after the epic. Every agent you run gets logged to it and Khint keeps a compact running summary, which rides along on the next agent if you tick the option. That is the context Atlassian cannot hold for you: it lives on your machine, spans the meeting app, the tracker and the wiki, and stops you restating the same product background ten times a day. Sessions stay in a local database and are never synced to Khint's servers.
Select the notes where the meeting left them
The notes are wherever you took them: a meeting app, a scratch file, a Slack draft, a mail. You do not move them into Jira first to get help with them. Highlight the block and press Cmd+Shift+K. The palette opens over the app you are in and lists the agents in your active pack. Pick “Notes to tickets” and the rewritten version comes back in place. No prompt written yet? In Agents, describe what you want in one line and press Generate, then Improve to sharpen it later without starting over.
The bug that arrived as a picture
A tester drops a screenshot of a stack trace and the error string is an image, so no AI reading your Jira project can see it. Capture handles it from the same shortcut: Extract text drags a region and puts the text on your clipboard, and Capture & ask opens a short conversation about what is on screen. The error message reaches the ticket without anyone retyping it off a picture.
File it with the fields your project enforces
Connect Atlassian and the palette gains a Jira Create issue row. Khint reads the project first to learn its issue types, its assignable people, its priorities and the custom fields it marks required, drafts the issue, then shows you one editable card. You change the summary, the type or a field and confirm, and the issue is created with its key and link coming back as the result. The write only happens after you press the button, so a draft you abandon costs nothing, and a required field is asked for on the card instead of surfacing as a failed request afterwards.
Publish the write-up, then save the pass
The decisions belong on a page, not in a ticket. A Confluence Create page row publishes the write-up, and Appendadds to a living page under a version lock that refuses to write rather than overwrite a colleague's concurrent edit. Once the sequence works it becomes one press: a workflow chains up to five steps, agent steps that draft and then integration steps that write, and carries its own keyboard shortcut. Next Tuesday the whole pass is that one key.
From the notes to both tools, in one run
Up to five steps in series, and each step's result becomes the input of the next.
The running text is what moves between steps: the agent's draft feeds the Jira write, and the Jira write's receipt, the issue key and its link, is what the Confluence page carries. Workflows can read from Atlassian too, not only write to it. A step can pull an issue, run a search or fetch the active sprint, and what it returns becomes the input of the next step, so “read the sprint, summarise it, post the summary” is one chain. The run panel shows each step ticking over with the line it produced, so you see what landed without opening a tab.
Where each one starts
Not a scorecard of models, a question of when in the day each one is reachable.
With Khint
- Starts from anything on screen: a selection in any app on macOS or Windows, a screen region, a document, or your voice.
- The instruction is a saved agent you reuse and share as a pack, so a team writes tickets the same way.
- Creates the issue, comments, changes status, publishes or appends a Confluence page, and reaches Linear, Notion, Slack and Gmail as well.
- Carries your own context: a Memory session summary and a local index of your Jira, Confluence, Notion and Linear.
- You download it and it works. Nothing to enable site-wide, and the session database stays on your machine.
With Atlassian Intelligence
- Built into Atlassian's cloud products and switched on for the whole site by an admin, so it is there the moment you open Jira.
- Works on Atlassian content: the issue you are reading, the page you are editing, the comment thread you need summarised.
- Right where the object is. No shortcut to learn and no connection to set up for the tools it already lives in.
- Work that is not yet an issue or a page reaches it the way it reaches any assistant, by pasting it in first.
- It knows your Atlassian site. It does not know the meeting app, the screenshot or the mail thread the requirement came from.
What Khint does not claim
The honest edges, so the comparison is worth reading.
It is not inside the Jira editor
Khint has no cursor in your issue description and does not suggest as you type there. It hands the finished text to Jira through the integration, as a create, a comment or a status change.
It reads what you hand it
An agent sees the text you selected, or the document you picked, plus your session summary when that agent has the option ticked. It does not watch your screen or your Atlassian activity feed.
The index is a local copy
Knowledge syncs your Jira, Confluence, Notion and Linear into a database on your own machine, so an agent can search it before drafting. It is a copy that syncs periodically, not a live view of the site.
Every write is deliberate
An integration step shows the draft on a card before anything is created, and Super Khint puts a Run button on each proposed action. Nothing is filed in your project while you are looking away.
You do not have to pick one
If the content is already an issue or a page, the AI inside Atlassian is the shorter path, and Khint is not trying to beat it there. The case for Khint is the half of the loop that never enters Atlassian until you put it there: the notes, the screenshot, the client mail, the Slack thread, and the twenty small passes a day that end in a ticket. The two sit side by side, one inside the site and one above every app on the machine. Worth checking with your security team first: agents only see what you select, any agent can redact identifiers before the text reaches the AI, and Memory never leaves your device.
Common questions
Is Khint an alternative to Atlassian Intelligence?
They overlap less than the question suggests. Atlassian Intelligence is built into Atlassian's cloud products and works on what is already there: the issue, the page, the comment thread you want summarised. Khint sits above every app on your Mac or PC, so it reaches the work before it becomes an Atlassian object, which is where a product owner's input actually starts. Select rough notes in a meeting app, press one shortcut, and a saved agent turns them into a well-formed issue that Khint then creates in your project. Most people who use both keep the built-in AI for reading what is already in the site and use Khint for getting things into it.
Can Khint create the Jira issue itself, or does it only draft text?
It creates it. Khint reads the project first to learn its issue types, its assignable people, its priorities and the custom fields it marks required, then drafts the issue and shows you one editable card. You adjust the summary, the type or a field and confirm; the issue is created and its key and link come back as the result. Because the card is shown before the call, cancelling costs nothing, and a required field your project enforces is asked for on the card rather than surfacing as a failed request after you already confirmed.
Does Khint work with Confluence as well?
Yes, one Atlassian connection powers both. The palette gains Confluence rows to create a page and to append to an existing one, plus adding a comment. The append is version-locked on purpose: Confluence has no partial update, so Khint reads the current page, adds only your new content, and refuses to write if someone edited the page in the meantime rather than overwriting their change. A workflow can end in a Confluence write the same way it ends in a Jira one.
Can a workflow read from Jira, not just write to it?
Yes. A workflow step can pull a specific issue, run a search, or fetch the active sprint, and what it returns becomes the input of the next step. That makes sequences like read the sprint, summarise it, then post the summary to a Slack channel a single chain behind one shortcut. Khint also keeps a local index of your Jira, Confluence, Notion and Linear on your own machine, which an agent can search before it drafts so a new ticket lines up with the ones already in the backlog.
Where does my text go, and does anything leave my machine?
An agent only ever sees the text you selected, plus the compacted summary of your active Memory session when that agent has the option ticked. Memory sessions and the local Knowledge index live in a SQLite database on your own machine and are never synced to Khint's servers. Any agent can be set to redact emails, phone numbers and similar identifiers before the text reaches the AI, and a team that prefers to run on its own account can bring its own Anthropic or OpenAI API key.
Keep the AI in Jira, add the step before it
Free with 300 credits a month, about 10 AI actions a day. No credit card, nothing to enable site-wide, no admin ticket to raise.