Frequently asked questions

Sharing a chat hands over the whole conversation, and the other person needs the same AI to carry on from it. With Kunobi each of you keeps your own AI and chooses the pieces that become shared work; your other chats stay private. Your AIs build on what you share, flag what doesn’t add up and remember what you decided.

No. You install one app, once. From there your AI works out what’s available and what applies. You set it up by talking to your AI, not by filling in a wizard.

You approve the project and who can join it. Your AIs work inside that boundary without asking you about every small update. Tools keep their own controls for spending, credentials and destructive actions.

You say yes once, and you can take it back whenever you like.

Yes, and that’s deliberate. Two sets of keys, two separate spaces, with nothing crossing between them unless you decide it does.

You share one project with the other person and both your AIs. Its research, files and decisions stay together. Only this project is shared: personal chats, unrelated files and credentials stay outside it. Decisions made there are kept, so both AIs can build on them the next time you work together.

What Kunobi keeps for you starts on your machine. Shared projects are available to their members; optional sync and backup have their own access settings.

What Kunobi keeps for you stays where it is and keeps working. It doesn’t need us to exist.

Claude, ChatGPT, Gemini, Codex and others, through the tool connections they already support. If a new one arrives next year, it works with that too.

Documents and local files. More over time.

The waitlist is. We’ll be clear about pricing well before anyone is asked to pay.

Kunobi is in development and we’re onboarding a small group first. Leave your email and we’ll be in touch when there’s a spot.

Things it needs to finish what you asked: a report, a dataset, a paid API call. Small, one-off purchases that would otherwise stop it dead. Not shopping.

Only up to the limit you set, and only on the kinds of things you allowed. Anything above that waits for you. You can pause or revoke at any point.

Usually cents. You pay for what fits: a few cents for a one-off, a subscription when something gets used every week.

Yes. When something gets used regularly, a subscription is the sensible answer. You approve it once, it counts against the same budget, and you cancel it in the same place you see everything else.

Every payment leaves a receipt showing what was bought, when, for how much, and what it was for. You can see the whole history and export it.

In a wallet on your machine, under a limit you control. Not in an account we hold, and the agent never has direct access to the keys.

No. You set a limit and top it up. Everything else happens underneath.

Yes. Separate budgets, separate limits, separate histories, with nothing crossing between them.

An open standard for machines paying for things. It uses the web’s “payment required” response, which existed for decades and was never used, so software can pay for one thing without creating an account first. AWS and OpenAI are building on it too.

Cards assume a person filling in a form, and a merchant relationship that lasts. Neither fits an agent that needs one report, once, at three in the morning.

Your remaining balance is yours and your receipt history is exportable. Nothing is locked in an account we hold.

Agent payments are in development. Leave your email and we’ll be in touch when there’s a spot.

Both AIs need a Kunobi connection to participate. The product plan also includes shared folders and read-only web previews, so an invited person can see the work before installing.

Only this project is shared. Personal chats, unrelated files and credentials stay outside it.

No. One of you can be on Claude and the other on ChatGPT, Gemini or Cursor. Kunobi sits underneath whichever each of you already uses.

No. A mission starts with two people. Anyone else joins when they’re invited, from whatever AI they already use.

Contradictions between what the mission’s members have asserted: a promise against a contract, a date against a dependency. It cites both sides and asks you to decide.

It replaces the part where you read documents to each other. What’s left is the decision, with the open points already on the table.

Only what you connect, and only for the agent of the person who connected it. The mission shares what its members put in (a spec, a contract or a decision), not their credentials.

A review happens after the work, on the diff. This happens as the work is planned: the other agent inherits the spec and challenges it while the date is still a proposal.

The same app has a deeper mode, Kunobi Ninja, with live clusters, storage and telemetry, for people who run infrastructure. It stays hidden until you turn it on.