Trust, earned

Agents earn trust.
None starts with it.

Five levels. Every agent enters at zero with no spending rights. Each level after that requires proof. Developer KYB, user delegation, thirty days of clean behavior, then institutional review.

Trust level progression
L0
Registered
L1
Verified
L2
Trusted
L3
Autonomous
L4
Institutional Enterprise only
5
Trust levels
Behavioral
Baseline fingerprint
Drift
Active detection
Auto
Demote on anomaly

Trust unlocks. Step by step.

Level zero is no spending. Each step above it unlocks only after the prior proof lands and is logged on chain.

01 / 05
L0

Registered

Starting state

The agent exists in the system. No financial operations are permitted. This is the starting state for every new agent until the developer completes attestation.

No payments No wallet access Read-only
L1

Developer verified

KYB attestation

The developer has signed a KYB attestation. The agent can transact within low spending limits on a single chain. Every transaction is recorded for behavioral baseline learning.

Low limits Single chain Basic payments
L2

User authorized

Human delegation

A human principal has signed a delegation that authorizes this agent to act on their behalf. The agent unlocks higher limits, multi-chain settlement, and access to the agent marketplace.

Higher limits Multi-chain Marketplace
L3

Behavioral baseline

Full autonomy

The agent has maintained thirty days of clean transaction history. The system has established a behavioral baseline and uses it to detect drift on every subsequent payment. Full autonomy within policy.

Full autonomy All chains Max limits
L4

Institutional principal

Enterprise only

Reserved for treasury and institutional agents under direct compliance officer review. L4 unlocks bundle-binding to a corporate legal entity, dedicated liability mapping, and the regulator-portal disclosure path. Promotion requires Veris L3 attestation plus an offline contractual review.

Treasury scale Compliance officer Regulator portal

Every agent. Live.

Trust level, behavior, and last activity, one query away. Drift detections and automatic demotions arrive the moment they fire.

02 / 05
Agent fleet
5 agents
procurement-bot
Cloud compute purchasing
L3 47 d clean
data-buyer
API data acquisition
L2 18 d clean
infra-scaler
Auto-scaling payments
L1 DRIFT
settlement-agent
Cross-chain settlement
L3 62 d clean
new-experiment
Testing phase
L0 Pending
1 drift alert active 4 / 5 healthy

Promote, suspend, or investigate any agent directly from the API or dashboard. Every state transition is signed and recorded in the activity log with the operator identity that triggered it.

Drift detections demote the agent atomically. The next policy evaluation sees the new level and applies the L1 ceiling. There is no propagation delay between detection and enforcement.

Drift. Caught.

Every payment is checked against the agent's normal pattern. Step outside, and the agent suspends inside the same cycle.

03 / 05

The behavioral baseline is a vector of features captured over the agent's lifetime: typical counterparties, spending cadence, asset mix, transaction size distribution, and chain preference. Drift is the distance between an incoming transaction and that baseline.

When drift exceeds the policy threshold, the agent is demoted in place. Demotion takes effect on the same evaluation that detected the anomaly. There is no human-in-the-loop delay. The developer is notified through the webhook subscription, and the incident is anchored into the L7 audit log for regulator disclosure.

Behavioral analysis · agent infra-scaler
Normal patternPASS
Within spending limitsPASS
Unusual destination clusterDRIFT
Agent demoted L3 → L1. Developer webhook fired.

People verify once. Agents verify every time.

KYC was built for humans. One check, lifetime trust. KYA was built for agents. Trust is rechecked on every payment.

04 / 05

KYC for humans

  • Verifies a human identity once at onboarding.
  • Produces a static approval that persists until a manual review event.
  • Requires government-issued documents and a selfie comparison.

KYA for agents

  • Verifies agent behavior continuously on every transaction.
  • Produces a dynamic trust score that increases on clean activity and decreases on drift.
  • Requires developer KYB attestation, user delegation, and an observable behavioral baseline.

Register. Promote when ready.

Create an agent at L0. Promote to L1 with an explicit call. Higher levels need verified proof. Trust never rises on its own.

05 / 05

L3 promotion is the only level that arrives without an explicit API call. The system evaluates the behavioral baseline after thirty days of clean activity and promotes the agent automatically when the threshold is met. Every other level transition is an explicit, signed action.

kya_example.py
# Register an agent at L0
agent = oris.agents.create(
    name="procurement-bot",
    description="Purchases cloud compute credits",
    kya_level=0,
)

# Promote to L1 after developer KYB attestation
oris.agents.promote(
    agent.id,
    target_level=1,
    attestation="kyb_signed_2026-05-26",
)

Ship safer agents today.

Register your first agent and set its trust through the API. KYA ships with every Oris plan. The behavior baseline starts the first time the agent pays.