Stop re-briefing your AIevery prompt
A focused task jumps across tools: a spec, a screenshot, a few tickets, a chat with Claude. Every time you switch, the AI starts cold and you re-explain the task. Khint holds one compacted work context for the whole session, so every AI action (and your AI tools) already know what you're working on.
The task is one thread,
your prompts keep forgetting it
Refining a backlog or writing a spec isn't one prompt. It's an afternoon of small moves across a handful of tools, all about the same thing. But each rewrite, each ticket, each chat starts from nothing, so you spend the day re-typing the same context. Khint fixes that with its three pillars, Agents, Capture, and Memory, wired together behind one palette, so the task's context follows you from move to move instead of resetting on every prompt.
One session, context that follows the task
From the first note to the last ticket, without re-typing what it was about.
Start a session when you go heads-down
You block off the afternoon to refine the Q3 backlog. Before you dig in, hit Start and name the session: “Q3 backlog · Wed”. From here on, Khint only pays attention while that session is on. There is no always-on capture and no keystroke log; it simply logs what you do inside the task and keeps a running summary as you go. How Memory works →
Khint compacts the task into a tight context
As you read a spec, paste a note, and draft a ticket, each entry lands on the session and Khint rewrites a short summary in the background: who, the subject, the decisions, the open items. You don't babysit it or write it yourself; it holds a paragraph or two, not a growing transcript, so the context stays sharp instead of ballooning.
Pull a fact off the screen without breaking flow
A number or an error you want to keep is sitting in another window. Open the palette with Cmd+Shift+K, run Extract text, and drag over the region, the text lands on your clipboard and is logged to the session. No AI rewrite, just the raw text, captured in seconds and now part of the task's context. How Capture works →
Every Action already knows the task
Run a saved Action: “Turn this into a ticket”, “Rewrite the acceptance criteria”, “Summarize for the epic”. Because the session is active, its compacted summary rides along on every prompt, wrapped in
<session_context>. No cold prompts, no re-explaining the task each time you switch from one Action to the next.Switch to Claude Desktop without re-briefing
You want to think a thorny epic through in Claude Desktop or Codex. Khint's bundled MCP server lets your AI client read the same active session live, read-only by default, no copy-paste. Ask it to draft the breakdown and it already has the task's thread; any write back into the session surfaces an explicit OS notification first. How integrations work →
Stop when the task ships: clean slate
The backlog is refined and the tickets are filed: hit Stop. The context closes and the next task starts clean, because Memory is short-lived on purpose. The session stays on your Mac for replay if you ever need to retrace what you decided. Refine a backlog with Khint →
Why not just use a long context window?
You can keep pasting into one long chat, but a raw transcript grows and drifts, the important decision from an hour ago gets buried, and it all sits in a chat history you don't control. Khint keeps a short rewritten summary instead of a wall of text: the active Memory session compacts as you work, attaches to every Action automatically, and is the same live session your AI toolsread over MCP. Because Actions only ever see the text you select plus that summary, and sessions never leave your device, the task stays on your Mac, and Stop wipes the slate so the next task doesn't inherit the last one's noise.
Common questions
What is a work session in Khint?
A work session is a short, explicit window of attention you turn on for one task: refining a backlog, writing a spec, working through a set of tickets. While the session is on, Khint logs every note you type, every Action you run, and every screen capture you take, and rewrites a tight compacted summary in the background: the subject, the decisions, the open items. Press Stop and the context is closed. It's about this task's work, not your whole week.
How does Khint keep AI context across different tasks and tools?
The active session's compacted summary is wrapped in <session_context> and attached to every AI Action you run while the session is on, so a rewrite, a summary, or a draft ticket already knows what you're working on without you re-typing it. To carry the same context into Claude Desktop or Codex (OpenAI), Khint ships an MCP server out of the box: your AI client reads the active session live, read-only by default, no copy-paste between tools.
Is this the same as a long ChatGPT context window?
No. A long chat window keeps a growing raw transcript that you paste into and scroll through, and it lives in a chat history you don't control. Khint keeps a short, rewritten summary (a paragraph or two, not a wall of text) and it's scoped to one task with an explicit Start and Stop. Actions only ever see the text you select plus that session summary you opted into. Nothing else leaves your Mac silently.
Where is my work session stored?
On your Mac. Sessions and their entries live in a local SQLite database and are never synced to Khint's servers, Memory is local-only. Your sessions stay on-device whether or not you're signed in. Stop the session and the context is closed; past sessions stay on-device for replay if you need to look back.
Try it on your next deep-work block
Free with 10 AI actions and 5 captures per day. No credit card. Start a session when you dig in, stop it when the task ships.