Describe the job,Khint writes the prompt
You know exactly what you want out of the AI: a raw requirement rewritten to your team's Definition of Ready, every time, the same way. What you do not have is an evening to spend learning how to ask for it. In Khint you type that sentence once and it drafts the whole agent, prompt included, for you to read and save.
You know the job,
not the phrasing
Every guide to using AI at work opens with the same homework: learn to write a good prompt. So the thing you actually wanted sits behind an evening of role prompts, few-shot examples and delimiters, a craft you will use twice and then forget. Khint takes the sentence you would have said to a colleague and turns it into a saved agent, with the prompt already written.
The palette comes to the text, not the other way around
No window to open, no tab to switch, nothing to paste twice. Khint reads the selection where it is and writes the answer back in the same place.
The real palette: the agent you generated runs on the selected text and the result is pasted back in place.
From one sentence to a saved agent
Building a “Definition of Ready” agent without writing a prompt.
Say what it should do, out loud
On the Agents page, open Generate an agentand type the ask the way you would brief someone: “rewrite a raw requirement so it matches our Definition of Ready”. That is the whole input. Khint drafts the agent from it: a name, a one-line description, and the prompt itself. Generating is free and does not touch your monthly credits.
It picks what the agent reads, too
A prompt is only half an agent. The draft comes back with an input source already chosen, from the four Khint offers: the text you select, a document, your voice, or your active Memory session. Ask instead for “fix the wording in my Word docs and give me the file back” and you get a document agent that returns the edited file with its formatting intact, not a wall of text to re-paste.
Read it before it exists
The draft opens in the agent editor, not in your pack. The prompt types itself out so you actually read it rather than scroll past it: rename the agent, cut a line that does not sound like your team, switch the input if it guessed wrong, then press Save agent. Khint validates the prompt at that moment and tells you plainly if it refuses one, so a broken agent never lands silently in your pack.
Run it, then press Improve
Select a rough requirement in Jira, hit Cmd+Shift+K, run the agent, read what comes back. Too terse? Reopen it: once an agent has a prompt, the Generate button becomes Improve and refines the existing prompt in place instead of starting from zero, so the parts that worked survive. Improving is free too. You iterate on the output, not on your prompt-writing skills.
Give it a key, or give it to the team
Record a shortcut in the editor and Khint tells you if the combination is already taken. Or leave it in the palette and run it on whatever is selected, in any app. When it earns its place, drop it in a pack and share the pack so five people stop keeping five slightly different versions of the same instruction.
Describing the job, or learning to write the prompt
Same goal, two very different evenings.
Describe the job
- One sentence, phrased the way you would brief a colleague. Khint drafts the name, the description and the prompt.
- The draft already knows what it reads and what it gives back: a selection, a file, your voice, your session.
- You read it before it exists. The agent lands in the editor as a draft, and only your Save writes it into the pack.
- A disappointing run is one Improve away, refined in place, and refining costs nothing.
- It ends as an object: a named agent in a pack, behind a shortcut you chose, ready to share.
Learn to write it yourself
- A blank box and an article about role prompts, few-shot examples and delimiters.
- You still decide by hand whether it reads a selection or a file, and wire the rest yourself.
- The wording that worked lives in a note, or in a chat history you scroll back through.
- When the output drifts you rewrite the whole thing, because you cannot tell which line caused it.
- Nothing to hand a teammate but a paragraph to copy, which they will edit into a sixth version.
What the generator will not do
It writes the instruction. It does not quietly do the job.
It does not read your work
The generator only ever sees the sentence you typed. Not your selection, not your files, not your Memory session. It is authoring an instruction, so it has no reason to look at the material.
It does not save behind your back
A generated agent is a draft sitting in the editor. Nothing is written into your pack, synced, or bound to a shortcut until you press Save agent, and you can throw the draft away.
It does not half-wire a voice agent
A voice agent has no palette row: its shortcut is the microphone key. So if the generator lands on one, Khint blocks the save until you record a combination, rather than saving an agent you could never trigger.
It does not spend your credits
Generating a prompt and improving one are both free. Credits are spent when the agent runs on real text, which is the part that does the work.
Or start from an agent somebody already wrote
Generating is not the only way past an empty pack. The Agents page opens on a gallery of starter packs: preview every prompt inside one, add it in a click, then rename what does not sound like your team. If a colleague has already built the agent you need, they share their pack and you clone it, with every prompt visible before it lands. The generator is for the day nobody has written the one you want. And when the job is not one step but four, the same idea builds a whole workflow from a description.
Common questions
Do I need to know prompt engineering to use Khint?
No. On the Agents page, open Generate an agent and describe the job in one sentence, the way you would brief a colleague. Khint writes the prompt for you, along with the agent's name and description, and you read and edit every word before it is saved. If you would rather not type anything at all, the Agents page also opens on a gallery of starter packs holding ready-made agents you can preview and add in one click. Prompt writing is something Khint does for you, not a skill you have to acquire before the tool is useful.
What exactly does "Generate an agent" produce?
A complete draft agent, not just a paragraph of text: a name, a one-line description, the prompt itself, the input source it should read (the text you select, a document, your voice, or your active Memory session) and what it hands back (text, or for a document, the edited file). It opens in the agent editor as a draft with the prompt typing itself out, so you can review it line by line. Nothing is written into your pack until you press Save agent, and Khint validates the prompt at that point and says why if it refuses it.
What if the generated prompt gets it wrong?
Reopen the agent and press Improve. Once an agent has a prompt, the Generate button becomes Improve and refines the existing prompt in place rather than replacing it, so the parts that already worked survive the pass. You can also edit the prompt by hand at any time, or duplicate the agent from the editor and keep both versions until one wins. Generating and improving are both free: neither counts against your monthly credits, so iterating costs you nothing but a few seconds.
Can it build an agent that reads a document or listens to me?
Yes, and it chooses that for you from the way you phrase the ask. Ask for an agent that summarises any spec you give it into five bullet points and you get a document agent. Ask it to fix the wording in your Word files and give the file back, and you get one that returns the edited document with its formatting intact rather than plain text. Ask for an agent that turns what you dictate into a clean Slack update and you get a voice agent, which Khint will not save until you record a shortcut, because that shortcut is its microphone key.
Where does the description I type go?
To Khint's proxy and the model that drafts the prompt, and nowhere else. The generator never sees your selection, your files or your Memory session: it only reads the sentence you wrote, because writing an instruction does not require the material. Later, when the agent actually runs, it sees the text you selected plus your active session summary if you switched that on for it. Memory sessions live in a local database on your machine and are never synced to Khint's servers.
Describe your first agent tonight
Free with 300 credits a month, about 10 AI actions a day. No credit card. Generating and improving prompts does not spend them.