HumanAuth

HumanAuth puts a human in the loop of AI agent workflows. When your agent wants to do something that matters — deploy to production, send money, delete data — it pauses and asks. You get a push notification, see exactly what was requested with every parameter, and approve or deny with Face ID. Every decision mints a signed receipt bound to the exact action you saw. Your backend verifies the receipt offline and rejects anything no human approved. No receipt, no execution. HOW IT WORKS • Your agent (Claude Code, Cursor, or any MCP-compatible tool, or the SDK) sends an approval request • The request fans out to your phone as a push notification • You review the action and its parameters — nothing is hidden • Approve with Face ID, or deny with one tap • A cryptographically signed receipt goes back; your systems verify it independently BUILT FOR TEAMS • Invite unlimited approvers to your tenant • Approval groups with quorum policies — require 2 of 3 approvers for the big stuff • Full audit trail of every request and decision • Time-boxed requests: approvals expire if nobody answers SECURITY FIRST • Biometric authentication on every approval • Ed25519-signed receipts your backend verifies offline • New devices require approval from an existing device • Report suspicious requests in one tap HumanAuth is for developers and teams running AI agents in production. A free tier of 1,000 approval requests per month is included — every security feature, no card required. Get started: pair your phone in minutes at humanauth.ai — six lines of code on the agent side.