Hello, Yelt.
Why we built a policy engine and audit ledger for AI agents that touch money — and what we mean by 'tamper-evident' when it's a CFO asking.
AI agents are starting to touch money. Not in the “maybe someday” sense — in the literal, this-quarter sense. Engineering teams are wiring Claude Code into Stripe. Ops teams are letting AR agents draft invoices and AP agents pay vendors. Treasury teams are asking whether the rebalancing bot is in fact rebalancing the right way.
The CFO and CISO have the same question: who is responsible when the agent gets it wrong?
Yelt is the answer. Two surfaces, one product:
- The dashboard. CFOs write policies in plain English. Yelt parses them into a typed runtime descriptor. Held actions surface in Slack. The audit ledger maps to FS AI RMF, OWASP, and SOC 2 controls.
- The CLI. Engineers run
npx yelt initand the next agent session inherits the org's policies. Sixty seconds.
What we mean by tamper-evident
Every audit event row in our ledger is SHA-256 hashed against the previous row's canonical payload. Postgres triggers reject UPDATE and DELETE on the audit table — even our own engineers can't silently amend history. The yelt audit verify CLI runs offline against a JSON export, so an auditor can confirm chain integrity without trusting Yelt at verification time.
A regulator-grade audit pitch is a cryptographic claim. We wrote the substrate first, the marketing later.
What's next
We're in private beta with a small set of design partners right now. If you run a team where AI agents touch money — pay invoices, post journal entries, send refunds, sign on-chain — email us at hello@yelt.ai and we'll talk.