Every release,real ones

What shipped, release by release. Written when the build lands, not after.

2026-07-27

Fix

Capture answers back, and Cancel really cancels

Reading text from your screen from the Capture page used to happen in complete silence: you'd pick a region and the app would sit there looking frozen until the text appeared. It now shows a “Reading…” state the whole time, and the window comes back before the read starts so you can actually see it. Cancel has been fixed properly, too. Before, it only hid the result — the read carried on in the background, then landed in your history and your work session anyway, sometimes reappearing over the page seconds later. Now cancelling genuinely stops the work, and nothing stray is saved. If the read had already gone through when you cancelled, Khint says so plainly instead of resurrecting the result. And the drop zone no longer advertises file types it can't read: it lists exactly what Khint accepts, and the size warning is gone since large images are shrunk for you.

“Extract text” on the Capture page now shows a “Reading…” state instead of looking frozen.
Cancel stops the read for real — no stray entry in your recent captures or your work session.
A read that had already finished when you cancelled says so, rather than popping back over the page.
The drop zone lists the image formats Khint actually reads, and drops the misleading size limit.

2026-07-26

Feature

Hold a shortcut, speak, and Khint types it out

Khint now understands your voice. Hold a shortcut, say what you want, let go, and the text appears right where you were typing — in any app. It all happens on your own machine: nothing is sent anywhere, there's no separate service, and it doesn't count against your daily limit. Khint even primes itself with your agent and workflow names, so things like project keys and product terms come out spelled the way you mean them. Voice isn't just for dictation, either: an agent or a workflow can be set to listen, so its own shortcut becomes hold-to-speak. And with the palette open, the same gesture fills the idea bar so you can read back what you said and press Enter, rather than firing off something unseen. Voice recognition needs a one-time on-device model; until it's in place, Khint tells you clearly instead of failing.

Hold a shortcut, speak, and the text lands at your cursor in any app.
Runs entirely on your device — private by design, no extra service, and it doesn't use your daily limit.
Khint primes itself with your agent and workflow names, so names and project keys come out right.
Point an agent or a workflow at your voice: its own shortcut becomes hold-to-speak.
With the palette open, dictation fills the idea bar so you can review it before running.

2026-07-26

Feature

Build workflows without hunting for hidden ids

Adding an integration step to a workflow used to mean typing in cryptic ids — a Jira issue-type id, a Confluence space id, a Linear team id — that you couldn't know without digging through a vendor's developer tools. These are exactly the fields the AI workflow generator leaves blank for you to fill. Now every one of them is a dropdown that lists your real projects, spaces, teams and issue types, drawn from the same place the palette already reads. Change the project and Khint clears the issue type that no longer belongs to it, so a workflow can't quietly fail at run time on a mismatched id after the earlier steps already did their work.

Integration steps in the workflow editor now pick their target from a list instead of asking you to type a raw id.
Covers Jira projects and issue types, Confluence spaces, Linear teams and more.
Changing a project clears an issue type that no longer fits it, so the workflow won't fail halfway through.
The same fields the AI workflow generator leaves blank are now the easy ones to fill.

2026-07-26

Fix

Memory and Capture stay dependable on long sessions

A batch of reliability fixes for people who lean on Memory and Capture all day. On a long work session, Khint's live summary could quietly freeze once you'd logged around thirty entries and stay stuck for the rest of the session — that's fixed, and the summary keeps a wide view of your work as it rebuilds. In Capture & ask, one failed question used to jam the whole chat until you closed the window; now it recovers on its own and you can simply ask again. Capturing a screenshot no longer leaves the image sitting on your clipboard in place of the text you'd copied when a capture is cancelled or hits your daily limit. And saving a Capture & ask conversation to Memory after a retry can no longer drop the question and keep only the answer.

Your session's live summary no longer freezes after roughly thirty entries on a long session.
A failed Capture & ask question no longer locks the chat — it recovers and you can ask again.
Cancelling a capture (or hitting your daily limit) no longer leaves a screenshot on your clipboard where your copied text was.
Saving a Capture & ask conversation after a retry keeps both the question and its answer.

2026-07-25

Feature

Keep your Capture & ask conversations in Memory

When you use Capture & ask to talk to Khint about something on your screen, that back-and-forth used to disappear the moment you closed the window. Now a “Save to session” button in the chat header writes the whole question-and-answer thread into your active work session, so the reasoning stays with the rest of your work and feeds everything Khint remembers. Ask a couple more questions and save again, and it only adds the new exchanges rather than duplicating what you already kept — the button even tells you when there's nothing new left to save.

A new “Save to session” button keeps a Capture & ask conversation in your active Memory session instead of losing it when you close the window.
Saving again after more questions adds only the new exchanges, so nothing gets duplicated.
The saved thread becomes part of what your agents can draw on later.

2026-07-25

Fix

A steadier palette, capture and first run

A round of fixes that make the everyday flow more dependable. Running two actions in quick succession could paste the previous action's result at your cursor instead of the new one — that timing bug is fixed. Cancelling a screen capture with Esc no longer leaves your clipboard empty for up to half a minute; your text comes straight back. The History shortcut and tray item now reliably open your recent actions instead of leaving you on whatever screen you were on. And the guided first run is more honest: if turning on Memory fails, it now says so instead of quietly doing nothing, the palette explains why it needs something on your clipboard, and stale hints were tidied up.

Chaining two actions quickly no longer pastes the previous result by mistake.
Cancelling a capture with Esc restores your clipboard right away instead of leaving it empty for up to 30 seconds.
The History shortcut and tray item reliably open your recent actions.
During the welcome tour, a failed “turn on Memory” step now tells you instead of silently re-arming.
Clearer palette messages, a working “/” to search, and an agent that reads your session or a document no longer leaves your clipboard overwritten.

2026-07-25

Fix

Limits and backups behave correctly behind the scenes

Two quieter fixes around your limits and your data. When you reached your daily limit while pulling text out of an image file, Khint used to fall back to your own AI key without a word and skip the upgrade prompt — now it shows you the same upgrade option you get everywhere else, and never quietly bills your own key at your limit. Separately, a background error could stop Khint from backing up your history to the cloud until you restarted the app; that can no longer happen silently.

Reaching your daily limit while extracting text from a file now shows the upgrade option instead of quietly using your own AI key.
History backups to the cloud no longer stall until an app restart after a background error.

2026-07-24

Feature

Point an agent at your work session or a document

Until now every agent read the same thing: the text you'd selected. Agents can now choose what they work on. In the agent editor, a new Input source control lets each agent read your selected text as before, your active Memory session — so your working state becomes the input directly — or a document. For documents you can either pick a file each time you run the agent, dropping it in, choosing a recent one, or browsing for it, or pin one file to the agent so it always uses it. Big files are summarised once behind the scenes so they don't slow the run down, while small files are read in full; when you run one of these agents, the palette shows where its input is coming from.

Each agent can now read your selected text, your active Memory session, or a document.
Point an agent at your active session and its working state becomes the input, with no selecting needed.
Document agents can ask you for a file each run (drop, pick a recent, or browse) or use one pinned file every time.
Large documents are condensed once so runs stay fast; small ones are read in full.
The palette shows where an agent's input is coming from when you run it.

Want to be in the loop?

No RSS feed yet. Drop us a note and we'll email you when there's something worth a release post.

Email me when there's news Try the latest build