#KiteBBW $KITE
There are two lazy ways to let an AI agent spend money.
Give it a standing key and hope, or approve every payment by hand and stop using the agent for anything real.
Kite's Passport sits in the gap between them, and what interests me is how many separate dials it gives the user.
Per the docs, an agent that wants to pay creates a spending session with a task summary, a cap per transaction, a cap on total spend, a time limit and an asset.
The user approves it once with a passkey, on any device that holds it.
After that the agent can spend up to the total without asking again until the session expires or the budget runs out.
The docs' own examples are small on purpose: $2 per transaction and $10 total over 24 hours, or $1 and $5 over two hours. The wallet is funded in advance, and payments show up in a dashboard and an activity log.
My read is that each dial covers a different failure. The per-transaction cap limits one bad decision, like an agent talked into a single huge payment. The total cap limits a slow bleed from a loop that keeps retrying.
The time limit handles access you forgot you GRANTED. The task summary ties the money to a stated purpose, which is what makes the log useful afterward.
A typical API key or saved card has few of these: no stated purpose, often no expiry, and no ceiling beyond the account balance.
A technical deep dive on Passport lays out the chain as human credential, user wallet, agent passport, session key, then the transaction, which keeps the agent a step removed from the user's own authority.
That's the right shape for an agent economy where software spends faster than people can review it.
The weak link is the human tap.
A session is only as safe as the person reading the summary and the caps before approving.
Will people actually read those fields, or will session approval turn into the new "accept all cookies" and quietly cancel out the safeguards built around it?
$KITE @KITE AI
There are two lazy ways to let an AI agent spend money.
Give it a standing key and hope, or approve every payment by hand and stop using the agent for anything real.
Kite's Passport sits in the gap between them, and what interests me is how many separate dials it gives the user.
Per the docs, an agent that wants to pay creates a spending session with a task summary, a cap per transaction, a cap on total spend, a time limit and an asset.
The user approves it once with a passkey, on any device that holds it.
After that the agent can spend up to the total without asking again until the session expires or the budget runs out.
The docs' own examples are small on purpose: $2 per transaction and $10 total over 24 hours, or $1 and $5 over two hours. The wallet is funded in advance, and payments show up in a dashboard and an activity log.
My read is that each dial covers a different failure. The per-transaction cap limits one bad decision, like an agent talked into a single huge payment. The total cap limits a slow bleed from a loop that keeps retrying.
The time limit handles access you forgot you GRANTED. The task summary ties the money to a stated purpose, which is what makes the log useful afterward.
A typical API key or saved card has few of these: no stated purpose, often no expiry, and no ceiling beyond the account balance.
A technical deep dive on Passport lays out the chain as human credential, user wallet, agent passport, session key, then the transaction, which keeps the agent a step removed from the user's own authority.
That's the right shape for an agent economy where software spends faster than people can review it.
The weak link is the human tap.
A session is only as safe as the person reading the summary and the caps before approving.
Will people actually read those fields, or will session approval turn into the new "accept all cookies" and quietly cancel out the safeguards built around it?
$KITE @KITE AI