The persistent customer
Most tools treat every session as a blank slate. The same agent shops across merchants, across weeks, across a customer’s family of agents. Identity follows the customer, not the checkout.
ABOUT KYA LABS
The trust layer for agentic commerce.
Agents aren’t bots. Four principles and a compounding trust graph built to prove it.
THE THESIS
What separates a trust layer from an auth layer.
Scores that compound
Shopper attractiveness, not authorization. Every trip moves the score in one direction or the other. Merchants get a gradient, not a gate.
Growth, not fraud
The score exists to unlock, not block. Most of this space is built on detection, which shrinks the market. Trust grows it.
The bureau precedent
Consumer commerce scaled because credit bureaus made strangers legible to merchants. Agentic commerce is the same problem, one layer up. Every agent is a stranger until something makes them legible.
Four principles to create a compounding trust graph that drives agentic commerce for retailers preparing for agent traffic.
THE COMPOUNDING TRUST GRAPH
Tag. Identify. Rate. Unlock.
Tag→
Radar
Watches merchant surfaces. Identifies autonomous agents. Feeds the graph.
Identify→
Badge
The agent’s persistent name. Designed to travel wherever the agent goes. MIT, open source.
Rate→
kyaScore
The longitudinal score. Designed to compound with every trip. 500–850, like a credit score.
Unlock
VerifAI™
Real-time scoring at the merchant edge. Under 100ms. Routes agents to the best path.
Successful shopping becomes the identity. Every trip makes the next one faster.
WHAT WE’VE SHIPPED
Evidence, not promises.
Badge SDK live on npm
@kyalabs/badge-sdk v1.1.1. Zero runtime deps, 140 tests, MIT licensed.
View on npm →UCP Credential Provider published
io.kyalabs.common.identity. Built to the UCP specification for agentic commerce.
View spec →1,500+ agent trips observed
Across 40+ merchants surveyed. kya labs research, Apr 2026.
See the data →WHERE WE’RE PUSHING
Open questions we’re working on.
Building the trust layer means answering questions the ecosystem hasn’t settled yet. Three we’re actively investigating:
Cross-merchant linkage without coordination.
How do you link an agent’s trips across merchants when each merchant only sees its own traffic? Most observability breaks at the domain boundary. We’re testing whether behavioral and cryptographic signals can close that gap without requiring merchant-to-merchant contracts.
Identity when agents fork.
When a managed-agent platform spins up a new agent from a template, what reputation travels with the new agent and what stays with the old? Score portability has no settled answer. Neither does score inheritance.
Reputation under adversarial conditions.
If the score creates the incentive, the score gets gamed. What reputation mechanisms hold up when an adversary can instantiate agents at zero marginal cost?
CONTACT
Get in touch.
Enterprise, investor, or partner, we’d like to talk.