Everything you feed it lands in it
Notes you type, screen captures you save, agents you run, the tickets they create, the files you add. A session is the trail of a piece of work, not a chat log, and it's written as you go.
Press Start when the work starts. Khint keeps a short brief of it and hands that to every agent you run, so you never explain the same context twice.
An agent normally runs on what you highlighted. Point one at the active session instead and it runs on the work itself: the tickets you filed, the decisions you made, what's still open.
Khint compacts the session into an objective, the open threads and the decisions, then hands that to the agent. Every line of the answer comes from an entry that's already in the session: nothing was pasted, and nothing was selected.
Not a transcript and not a chat log. A session is the trail of one piece of work, plus the short brief Khint keeps rewriting from it, plus a recap written for you rather than for a model.
Notes you type, screen captures you save, agents you run, the tickets they create, the files you add. A session is the trail of a piece of work, not a chat log, and it's written as you go.
One line at the top of the session. Khint folds it into everything it keeps, so the brief stays on the goal instead of drifting with whatever you did last.
The brief is for your agents. The Recap tab is for you: plain language, written on demand, with one button that copies the whole session ready to paste into a doc or a message.
Khint rewrites the brief as entries land, and rebuilds it from scratch as the session grows. Objective, open threads, decisions, notes: one page, whatever kind of day it was. The What agents see tab shows it to you, exactly as an agent receives it.
Memory isn't a recorder you forget about. It starts on a press, holds one piece of work, and the whole of it sits in a file on your machine.
Security and privacyA session knows this morning. Knowledge knows the last two years: connect Jira, Confluence, Notion or Linear and Khint keeps a local full-text index of them, so an agent that needs the spec finds it in milliseconds, without a model call and without a copy of your corpus on anyone's servers.
A brief nobody reads is a diary. This one is wired into the three things that act on it.
The session you're in is a switch on the agent itself. Your indexed tools need none: once a source is on, every run reads what matches.
See what an agent isSet a workflow's input to the active session and one shortcut turns your week into an update, a ticket, a status page.
See workflowsKhint ships an MCP server: any MCP client can read the session you're in, no second briefing.
See the integrationsMemory is the context. These three are what use it.
Start a Khint Memory session before the task. Khint compacts what lands in it as you work, and every agent you run receives that brief automatically. To pipe it into Claude Desktop or Codex, install the bundled Khint MCP server: your AI client reads the active session live, with no copy-paste.
A compact working state Khint maintains for you: the objective, open threads, decisions, artifacts and notes, rewritten as entries land and rebuilt from scratch as the session grows so it never bloats. The session page has a What agents see tab showing the exact text your runs receive. You can go further and set an agent's input source to Active session, which makes that brief the whole input: the agent answers about your week with nothing selected.
No. Memory only keeps what happens while a session is on, and only inside Khint: notes you write, agents you run, captures you save, files you add, writes to your tools. No screen recording, no keystroke logging, and nothing at all before you press Start.
Rewind records your screen continuously in the background and searches that recording later. Khint listens only between Start and Stop, holds one piece of work at a time, and its whole job is to hand that work to the next AI run. Sessions live in a local SQLite file on your Mac and never sync, on any plan.
Yes. Beside the compact brief, every session has a Recap tab: a plain-language narrative generated on demand, plus a Copy recap button that puts the whole session on your clipboard, ready to paste into a doc or a message. Writing a recap costs no credits.
Knowledge is a local full-text index of the work tools you connect (Jira, Confluence, Notion, Linear), kept on your Mac and never on our servers. Indexing makes no AI calls and retrieval is local search that answers in milliseconds, so it costs no credits. Once a source is indexed, every agent run searches it and reads the few excerpts that match your text, and when nothing matches well nothing is added.
Free with 300 credits a month. No credit card. macOS 13+ or Windows 10 22H2+.