How it works
What happens between your agent wanting something and having it.
Your agent asks for something. It gets a price instead. Pocket checks that price against your limits, and only then does a wallet sign for it.
The whole exchange takes a couple of seconds and needs nothing from you.
Step by step
Your agentPocket
asks to buy something
PocketSeller
requests it
SellerPocket
answers with a price instead
checks every limit you set
If any limit fails, it stops here and the refusal is recorded.
PocketWallet
asks for a signature, only if allowed
PocketSeller
presents the signed payment
SellerPocket
takes payment and returns the data
PocketYour agent
hands over what it asked for
The decision sits between the price and the signature
Why the order matters
Pocket decides before anything is signed.
A refused purchase has no signature attached to it, because one was never created. There is no signed payment sitting around that a clever prompt could talk your agent into using.
When a purchase is refused
Your agent is not just told no. It is told what stopped it, and how much room it had left.
Refused Full market report
Amount 0.75 USDC
Rule per-job budget
Left 0.03 of 0.50A good agent reads that and picks something cheaper, instead of retrying the same request until your patience runs out.
When you want the final say
Set an approval threshold and anything above it waits for you. You approve or reject it from the payments screen.
Waiting for approval is not a soft yes. A held purchase has no signature, and the wallet is never asked for one until you release it.
What your agent can never do
- Read your API key
- Change its own budget or rules
- Pay for something Pocket refused
- Sign anything itself