Three modes. One model.
Verify.
Two-step identity verification. Photo ID, live face, secondary proof. Behavioral signals scored in real time. Tamper-resistant by design. Auditable manifest at session close.
Observe.
Rear camera. Real-time labeling of objects, scenes, faces, and readable text. A structured ledger of the physical world, built frame by frame, exportable on demand.
Narrate.
Front camera. Continuous narration, OCR, micro-expression tracking, scene tagging. Used wherever understanding the person on camera matters as much as what they say.
Built for industries where capture has to be proof.
If a claim, a contract, a credential, or a payout depends on what was actually captured, YUBI produces the record that holds up after the fact.
Cut claim cycle time. Replace paperwork-and-attestations with a signed video record of the loss, the ID, and the proof-of-loss artifacts.
Pass audits without re-running the session. Every verification ships with a SHA-256 manifest, frame ledger, and behavioral signals — exportable to your auditor.
Run thousands of sessions a day without standing up a vision team. One API, one model, three modes — drop it into the workflow you already have.
Not sure if YUBI fits your workflow?
Describe it. Get a real answer.
One or two sentences. YUBI's qualifier reads your use case against what the platform actually does and returns a fit decision with reasoning — no sales call required.
Two ways in. Same model.
API.
Drop YUBI into your stack. Structured JSON in, structured JSON out. One endpoint per mode. Stream frames, get a signed record back.
POST /v1/sessions { "mode": "verify", "claimed_name": "Jane Doe" } → 200 OK { "session_id": "ses_2k1f...", "stream_url": "wss://api.yubi.ai/...", "expires_in": 600 }
Dashboard.
A white-labeled portal where your teams run sessions, review records, and pull manifests. No build required.
- Session Replayv1
- Cost Attributionv1
- Tenant Isolationv1
What you get back.
Every session produces a structured, signed record. Frames, observations, behavioral signals, and outcomes are bound together with cryptographic hashing — so the record you get back is the record that was captured. Nothing edited, nothing inferred after the fact.
{
"session_id": "ses_2k1f8h3a",
"mode": "verify",
"outcome": "verified",
"trust_score": 0.94,
"steps": 2,
"frames_captured": 148,
"duration_sec": 47,
"manifest_hash": "sha256:9f2a...e1b8",
"completed_at": "2026-05-25T14:22:09Z"
}