Point an agent at a file,not just your selection
Half of a product person’s raw material never sits selectable on screen: it arrives as files. Transcript exports, feedback CSVs, Markdown specs. Khint agents can take a document as their input: pick the file at run time from an inline picker, or pin one file to the action for good. No opening, select-all, copy, paste ritual.
The input is a file,
the workflow shouldn’t change
Select text, hit the shortcut, pick an action: that loop works beautifully when the text is in front of you. But a one-hour interview transcript is a file in your downloads folder. So is the CSV your support tool exported, and the spec a colleague sent over. The usual workaround is to open the file, select everything, copy it into a chatbot, and paste the answer back: exactly the copy-paste tax Khint exists to remove. With the Document input source, the agent reads the file itself: you pick it (or pinned it once), the agent runs, and the result comes back as plain text. Same shortcut, same palette, one less detour.
From a file on disk to a finished result
Choose the source, pick or pin the file, run.
Tell the agent where its input comes from
Every agent has an input source, set with a small switch in its editor: Selected text (the default), Active session (your Memory working state becomes the input), or Document (a file becomes the input). Flip an action to Document and it stops caring what you have highlighted: the file is the input.
Pick the file at run time…
With Ask me each run, launching the agent from the palette first opens an inline file picker: a drop zone, your eight most recent files, and a browse button for anything else. Drop this morning’s transcript export on it, and the agent runs on that. Tomorrow, a different file, same agent. Cancel the picker and nothing runs, nothing is charged.
…or pin one file to the action
With Always use this file, you choose the file once, in the editor, and the action is bound to it. That fits documents that live at a fixed path and evolve: the running glossary your team keeps as Markdown, an exported requirements sheet you refresh weekly. Run the action and it reads whatever that file says today.
Long files are condensed once, then cached
Small files (up to 10KB) go to the agent raw. Bigger ones are condensed into a compact digest the first time, and the digest is cached on your machine, tied to the file’s exact content. Re-run on an unchanged file and the cached digest is reused instantly; edit the file and the digest is rebuilt. The condensing never counts against your daily actions.
The output lands like any other action
From there it is a normal Khint run: the agent’s prompt does its job on the document’s content, and the result comes back where you work. Summarize the transcript, extract the open questions, turn the feedback CSV into draft backlog items: the file was the input, but the output is ordinary text you paste wherever it is needed.
A BA’s week, file-first
Monday, three user interviews land as transcript exports. Your “Synthesize interview” agent is set to Ask me each run: hit the shortcut, drop the first file on the picker, and get themes and open questions back; the next two transcripts are one run each, sitting right there in the recent-files list. Wednesday, backlog triage: the support CSV goes through a “Feedback to backlog items” agent the same way, then the draft items flow into your usual feedback-to-backlog routine. And all week, a “Check against glossary” action stays pinned to the Markdown glossary your team maintains: it reads whatever the file says the moment you run it, so the answer tracks the document, not a stale copy. For interview work specifically, see user interview synthesis.
Document input vs Memory sessions
Khint has two ways to hand a file to the AI, and they answer different questions. Dropping a file into an active Memory session makes it part of your working context: every agent that opts into session context sees a compacted summary of it, alongside your notes and pulled tickets, for the rest of the session. The Document input source is the surgical version: one agent, one file, one run, with the file as the entire input rather than background. Reach for Memory when the document should color everything you do this afternoon; reach for Document input when the document is the task. They compose, too: an agent running on a document can still opt into session context, so a transcript summary can come back already aware of the sprint you are in.
Common questions
What kinds of files can an agent read?
Plain-text files: meeting transcript exports, Markdown specs, CSV exports, log files, code, anything that opens as text. Khint reads up to 50KB of the file (longer files are truncated at that point, with a marker so you know). Binary formats like PDF or Word documents are not parsed: export them to text first.
Do long documents burn my daily quota on every run?
No. A file of 10KB or less is passed to the agent as-is. A longer file is condensed once into a compact digest, and that digest is cached on your machine keyed to the file's content: rerunning the same agent on the same file reuses the cache with no extra AI call, and the condensing step itself does not count against your daily actions. Only when the file's content changes is the digest rebuilt. The agent run itself costs one action, like any other.
How is this different from dropping a file into a Memory session?
A Memory session file drop feeds the session's working state, so every agent that opts into session context sees a compacted summary of it alongside whatever else you logged. Document input is narrower and more direct: one agent, one file, one run. Use Memory when a document is background for everything you do this afternoon; use Document input when the file is the thing the agent should transform right now.
Can a workflow step read a document too?
Not yet. Document input applies when you run the agent directly from the palette or its shortcut. Inside a workflow, each step receives the running text from the step before it, so a chain still starts from selected text. If you want a document to influence a workflow, drop it into your active Memory session instead.
Your files are inputs now
Free with 10 AI actions and 5 captures per day. No credit card. Set an agent’s input to Document and run it on the next file that lands in your downloads.