We build authorization for AI agent payments
Saifuro checks each payment an AI agent proposes against the mandate its operator set, and signs the answer so the counterparty can verify it. It never holds funds. Here is who builds it and how we work.
Why we're building this
Software is starting to hold budgets. Agents buy data, rent compute and pay each other over rails that launched within the last two years. Each protocol brings its own wallets and receipt formats, and none of them covers spending limits, escrow or audit.
That layer should be neutral, so Saifuro does not operate a rail of its own and does not hold customer funds. It checks your rules on whatever protocol a transaction takes, signs every decision, and records it in a log an auditor can read.
The team
Five people lead product, engineering, marketing, finance and recruiting. They are part of a larger team, and we are hiring to grow it further.
How we operate
Non-custodial by design. Your funds stay in your accounts and wallets. We check each payment against your rules and sign the answer; your own systems act on it, and we never hold the money.
Policy before payment. Each payment request is checked against your limits before it is paid. In enforce mode your systems stop a request that breaks a rule, and one above its review threshold waits for your approval.
Open standards first. Verdicts are JSON Web Tokens signed with ES256 and checked against keys we publish, and a verdict travels beside an x402, ACP or AP2 payment rather than replacing it.
Everything auditable. Every decision is written to an append-only log with its signed verdict. The log exports as NDJSON, ready to hand to an auditor.
Official channels
These are the accounts we run. Announcements land on X, LinkedIn and Telegram first, longer posts go to Medium.




