Your AI tools read the work session you're in right now. And the palette reaches into Jira, Confluence, Linear, Notion, Slack and Gmail, pulling work in and writing results back without you opening any of them.
Every call leaves from your own machine, and each row is one you switched on yourself. The AI tools section is the other direction: your assistant reading the same session over MCP.
Some tools read Khint. Khint reads and writes the others. Each one has a page of its own with what it adds to the palette, how to connect it, and what it will refuse to do.
Khint is an MCP server. Your assistant reads the session you're working in, and can write back into it.
Ask Claude what you're working on and it answers out of your live Khint session, with no pasting.
One config file covers the Codex CLI, the desktop app and the IDE extension. Same session, same five tools.
The khint-mcp binary ships inside the app. Point any MCP-compatible client at it and you get the same five tools.
Connect a tool once and its rows appear in the palette: pull an issue, a page or a thread into your session, and file the result back without leaving the app you're in.
Pull an issue, your active sprint or a JQL search into your session. Create issues, comment, change status.
Pull a page into your session. Publish a new one, comment, or append to an existing page without touching what's already there.
Turn anything you've selected into a Linear issue with a written title and description, filed in your default team.
Create a page in your database, append to an existing one, or pull a page and a search into your session.
Post an update to a channel. Pull a channel, a thread or a search back into the session you're working in.
SoonDraft or send an email from what you selected, and pull a thread into your session. Sending always asks first.
One companion that talks to an app on your machine rather than to an API.
Four things, and none of them is a new window to visit. The rows land in the launcher you already open, what you pull or file lands in your work session, the same writes are available to a workflow, and switching a card off keeps your credentials.
A connected tool doesn't add a window to Khint. It adds a section to the palette, and you pick which rows appear: the primary write is on, the reads are opt-in, because a launcher showing twenty rows on day one is a launcher nobody reads.
A pull isn't a viewer and a write isn't an export: both ends meet in the work session Khint is already keeping. The Slack thread you pulled is what the agent drafts from, and the ticket it filed goes back in as a receipt your AI client reads next.
Every write is also a workflow step, and a parallel step dispatches two to four of them concurrently from the same running text. Branches are best-effort on purpose: a message that landed can't be un-sent, so only a total wipe-out is a failure.
Turning a card off hides its rows and stops workflows using it, while the credentials stay in your keychain for an instant reconnect. Forget credentials is a separate, deliberate two-click button.
A hosted automation platform has to hold your credentials, because it calls your tools while your laptop is shut. Khint calls them while you're there, which removes the reason to hold anything.
Security and privacyConnect a work tool, plug in your AI client, or put a write inside a workflow. Each one takes a couple of minutes.
A personal token on your own account, and the palette gains that tool's section.
One toggle writes the client's config file. Restart it and your session is a tool it can call.
How the server worksThe same writes are steps, so a routine can end in your team's tools without you opening them.
See workflowsIntegrations are where the work comes in and goes out. These three are what happens in between.
No, and that's the whole design. Every credential goes into your operating system's keychain and every call goes from your Mac straight to Jira, Slack or Notion. Khint's backend never holds a token, never proxies an integration call, and has no connector infrastructure to breach. A hosted automation platform works the opposite way: it stores your credentials so it can call your tools while your laptop is closed.
Your Khint work sessions, and only those: the list of them, one by id, the live compacted state of the active one, and a search across their entries. Claude Desktop and Codex don't get a Jira tool or a Slack tool. If they see a ticket, it's because a palette row pulled it into your session first.
Not by default. A pull lands in a local Memory session, which lives in a SQLite file on your Mac and is never synced. If you switch Knowledge on for Jira, Confluence, Notion or Linear, Khint keeps a full-text index of those tools, still on your machine and still never on our servers. Turning a source off deletes its copy on the spot.
For Jira, Confluence, Linear and Notion, no: you create a personal token or an internal integration on your own account and it carries your own permissions. Slack asks for a small app of your own, which some workspaces require an admin to approve. Claude Desktop and Codex need nothing at all beyond Khint itself.
The API calls are free: pulls, searches and every picker make no AI call at all. What costs is the writing, when a row hands your selected text to an agent to turn it into an issue, a page or an update. That's one normal agent run against your monthly credits, the same as any other action.
It pauses. The rows leave the palette, workflows stop using it, and the credentials stay in your keychain for an instant reconnect. Clearing them is a separate, deliberate two-click Forget credentials button, and for the four indexed tools it also deletes their local Knowledge copy.
Free with 300 credits a month. No credit card. macOS 13+ or Windows 10 22H2+.
Create issuewrite
Create pagewrite
Pull threadread