An hour of talk in,the action items out
Meeting notes bury the three things someone actually agreed to do under a dozen decisions and tangents. Khint runs an AI Agent on the notes you already typed and returns just the action items (owner, task, due date) from one keyboard shortcut.
The commitments are in there,
they're just buried
A real meeting is mostly noise around a few commitments. By the time it ends, the three things someone agreed to do are mixed in with decisions the group made, status updates, and a roadmap tangent that went nowhere. Re-reading the whole wall to fish out who-owns-what's the chore that gets skipped, and skipped action items are the ones that quietly don't happen. Khint closes that gap: an Agent you tune once pulls the commitments out and drops the rest, Memory keeps the meeting's names and context attached to the draft, and when an item earns a ticket the integrations file it without leaving your notes.
Noise in, a checklist out
The same notes, before and after one agent.
- Maya to redo onboarding copy by Thu
- Agreed export bug is P1
- Long tangent on the roadmap…
- Sam will book user-test sessions
- Someone ping Legal (Maya?)
- ☐ Maya: redo onboarding copy (Thu)
- ☐ Sam: book user-test sessions
- ☐ Maya: ping Legal
One meeting, start to a clean list
A weekly product sync, step by step.
Start a session for the meeting
Open Memory and start a session: “Weekly product sync”. Every agent you run is logged and Khint keeps a compact running summary, so when you extract action items the model already knows the names and the topic that came up earlier in the call. How Memory works →
Save an extraction agent once
In Agents, create an agent whose whole job is to surface commitments: “return only the action items as a checklist (one line each, owner first, due date if it was said) and drop decisions, status updates, and side chatter.” Describe it in plain language and let the editor Generate the prompt, then Improve it as your meetings teach you what to filter out.
Type the meeting as it happens
You keep your usual scratch notes: “Maya to redo the onboarding copy by Thu. We agreed the export bug is P1. Long tangent on the roadmap. Sam will book the user-test sessions. Someone should ping Legal (Maya?)” A wall of decisions, asides, and half-assigned to-dos, exactly as it came out of the room.
Run the extraction in place
Select those lines and trigger the agent from the palette (Cmd+Shift+K). It returns just the commitments: “Maya: redo onboarding copy (Thu)”, “Sam: book user-test sessions”, “Maya: ping Legal”, with the export-bug decision and the roadmap tangent left out. The clean list lands where your cursor is, ready to paste into the meeting thread.
Turn an item into a ticket when it earns one
When a single item deserves to be tracked, the palette can file it: with Jira or Linear connected, a Create issue row drafts the ticket from the selected line and hands back the issue key. And every agent you ran is in your Memory session, so the meeting reads back as a record of what was committed and what you turned into work. How integrations work →
Extraction is just the first step
The action-item list is reusable text: paste it into the meeting thread, drop it into a status update, or feed a single line into a follow-up agent. Build a workflow of up to five steps in series and the list can roll straight into a filing step: the same machinery writes to Jira, Confluence, and Linear, so a sibling flow can publish the full minutes while the items themselves become tickets. And because Khint ships an MCP server, Claude Desktop or Codex can read the meeting session directly (the notes, the extracted items, the tickets you filed) without copy-paste.
Common questions
How does Khint extract action items from my meeting notes?
You write an AI agent once ("read these notes and return only the action items as a checklist, each line with an owner and a due date if one was mentioned; ignore decisions and side chatter") and save it in a pack. Then you select your raw notes anywhere and run that agent from the palette with a keyboard shortcut. The model returns just the action-item list, in place of the noise. The agent editor can Generate that prompt from a plain-language description, so you're not writing prompt boilerplate by hand.
Does it tell the action items apart from decisions and discussion?
That's entirely down to how you phrase the agent prompt, and the phrasing is the whole point. A prompt that says "return only commitments somebody agreed to do, not decisions the group made and not topics that were merely discussed" gives you a tighter list than a generic summary. Because the agent is reusable, you tune that wording once and every meeting after gets the same discipline.
Can it remember who owns what across a recurring meeting?
If a Memory session is active, the meeting's context rides along with each agent. Start a session for the meeting and everything you run is logged and compacted into a short running summary that's attached to the next agent, so when you extract items late in the call, the model already knows the names and the topic from earlier. Sessions are stored locally in SQLite on your machine and are never synced to the cloud.
What does Khint send to the AI, and where do my notes go?
Only the text you selected, plus, if a Memory session is active, the compacted session summary. Nothing else leaves your device silently. Khint runs on the free tier at 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 your next meeting
Free with 300 credits a month, about 10 AI actions a day. No credit card. Select the notes, run the agent, keep the action items.