Lares watches the agents that can touch your most valuable systems. A product in that seat has to be boring in the right ways: locked down by default, honest about what it stores, unable to surprise you, and reversible when it acts. This page is the plain-language version of how we hold ourselves to that.
Nothing here is aspirational copy; each one is how the product behaves today, and most of them you can check from the console yourself.
No self-serve signup, no open registration. Google and Microsoft single sign-on or an email magic link, with invites that can be bound to a verified address so a forwarded link grants nothing.
Roles are enforced server-side: owners act, viewers read, and workspace access can be scoped down to individual product modules, with seat caps on invited teammates.
Every organization runs in its own isolated world: fleet, findings, policies, notifications and audit trail. Nothing is shared across tenants, and one tenant's identifiers resolve to nothing in another's.
Log exports you upload are parsed in memory and discarded; the raw file is never written to disk. Agent credentials are fingerprinted with a one-way hash, so Lares can recognize a key it has seen without ever storing the secret itself.
Durable state lives in per-tenant snapshots on infrastructure you control.
The decision log is append-only, and every record is hash-linked to the one before it, so an edit, a deletion or a reorder breaks the chain visibly. The chain can be re-verified on demand, exported as JSONL, and streamed to your SIEM as it happens.
Containment fires beside your agents, never inside their request path, and every action has an undo that is tested before the action is ever offered. By default a named human confirms every verdict; Lares acts alone only where you explicitly dial it, per asset class, with thresholds.
Every outside integration, SIEM, email, ticketing, identity provider, LLM assistance, is opt-in and off until you configure it. Unconfigured, Lares runs fully self-contained: no traffic leaves to any third-party model or service.
The console badges every integration seam honestly: fixture,
dry-run or live, so you always know what is real.
Lares is a design-partner-stage product, and we say so. We are not yet SOC 2 or ISO 27001 certified, and until an independent auditor says otherwise you will not see those badges here.
What we can honestly claim: the architecture is built against those frameworks' control expectations from day one, least-privilege access, role separation, tenant isolation, an append-only tamper-evident audit trail, and encrypted transport, so the certification path is paved, not a rebuild.
For AI-governance frameworks (EU AI Act, NIST AI RMF, ISO/IEC 42001, NIS2), see the compliance crosswalk: where Lares produces the evidence those frameworks ask deployers of AI for.
If you believe you have found a security issue in Lares, tell us before you tell anyone else, through your Lares contact or the access-request channel on the front page. We acknowledge quickly, we fix before we publicize, and we will credit you if you want the credit.
We extend the same courtesy in reverse: if an issue ever affects your tenant's data, you hear it from us directly, with the audit record to reconstruct exactly what happened.
The fastest audit of a vendor is watching the product behave: the invite gate, the roles, the tamper-evident record, the tested undo. All of it is on the live console.