Fingerprint
Structured intent + constraints become a stable hash — not a vague “yes” to whatever ships later.
Human approval · Execution control
Secure agent actions at the moment of impact — then prove execution matched what was approved.
AgentOG fingerprints high-impact requests (money, bookings, data, commitments), collects explicit human approval for that payload, mints a short-lived token, and blocks execution when reality drifts.
Voice hits your deployment via AgentPhone webhooks when configured — same approval envelope as the web demo.
How it flows
- 1Agent proposes a concrete action
- 2Fingerprint captures payload + policy context
- 3Human approves that exact action
- 4Short-lived token issued
- 5Gate allows only a matching execution
Product pillars
Approver
Email, SMS, voice, or your own channels — the human sees vendor, amount, terms, and sensitive-field rules before committing.
Execution gate
Runtime checks token + canonical payload. Divergence → deny and redo with a corrected bundle (e.g. Stripe only after match).
MFA proves identity · AgentOG proves action security
Typical MFA
Confirms the human at the keyboard or phone.
AgentOG
Confirms the precise AI-initiated action that person approved — if the agent swaps vendor, price, or timing, execution stops.
Try the live story
The dashboard runs one vertical end-to-end (voice webhook → planning → optional live browse → approval → gate). The ride line is demo only; the same envelope applies everywhere below.
- Ride
- Pharmacy
- SaaS purchase
- Appointments
- Forms
- Refunds
- Procurement
- Caregiving
Book a cab from 560 20th Street to Ghirardelli Square after 5 PM, under $50, with wheelchair assistance.