Guarantees the shape of the system makes true.
An agent can run for hours and send sensitive prompts, tool results, and internal context through its model API. The boundary around that inference traffic has to be structural, not a promise in a policy document. This page explains what Layer X1 guarantees today, in plain terms.
Separation you can't misconfigure.
The strongest controls are the ones nobody has to remember to apply. Tenancy comes from the credential rather than the payload, and isolation is baked into the storage key itself — so there is no filter to forget and no flag to set wrong.
The drawing on the right is the whole mechanism: three requests arrive, the one asserting a tenant its key didn't earn dies at the gate, and the two that pass land in one physical pool as salted cells that can never address each other.
Your key decides your tenancy
Which tenant a request belongs to is derived from the credential that authenticated it — never from anything in the request body. A payload that tries to assert a different tenant is rejected at the door, not reconciled later.
Caches can't cross tenants
Every cache and storage key is salted with an isolation domain derived one-way from your authenticated tenant. Two tenants cannot address the same entry — separation is a property of the key, not a lookup that has to remember to filter.
Provider credentials are never yours to hold
You hold one Layer X1 key. Upstream provider credentials live with the gateway, rotate with the gateway, and are never issued to you, printed in a response, or exposed to your agents.
Keys are named, and revoked instantly
Issue a key per agent, per machine, per environment — each one named, so months later you can tell them apart. Revoke from the dashboard and it stops authenticating; nothing else you run has to change.
Passed through. Not collected.
Traffic is the product, not the asset. What we keep, we keep briefly and for reasons we can name — running the service, finding failures, stopping abuse. The full detail, including your rights over it, is in the privacy policy.
Nothing trains on your traffic
The content of your requests and responses is not used to train models — not ours, not anyone's. Serving traffic is serving traffic; it is never retained as a dataset or sampled for evaluation.
A short window, then gone
Request metadata and request/response content are kept for a limited period — up to 30 days — for reliability, debugging and abuse prevention, then deleted. Account data lives as long as your account does.
We never see your card
Payment details are collected and held by our merchant of record. They do not pass through our systems and we do not store them.
A short list of processors
Payments, authentication and database, and hosting — that is the list. Your data goes nowhere else, except where the law requires it, and it is never sold.
Found something? Tell us first.
If you believe you have found a vulnerability, email security@layerx1.com with enough detail to reproduce it. We will confirm receipt, keep you updated while we work, and credit you when it is fixed if you'd like us to. Please give us a chance to ship a fix before disclosing publicly.
One key, one boundary, and inference built for long-running agent work.
Running this inside your own network instead? Sideren is the private-deployment surface, where the boundary is your perimeter and nothing egresses at all.
Your agent stays the same.
Its inference gets better.
Start free · no card · Starter from $5/mo