Design partner pilots · H2 2026
Audit-grade, verifiable, client-controlled records of what your AI agents actually did, credible to the regulators, courts, and insurers.
The problem
Enterprises are moving from small AI pilots to hundreds of AI agents taking real action at machine speed. When a regulator, court, or insurer asks what your AI did, most teams have a log: plain-text rows, editable by anyone with access, rotated on a retention schedule.
Any system can show that an agent acted. Few can prove what it did, who authorized it, and that the record hasn't been altered since. Here's what's missing when the subpoena arrives:
With VAiDR, each of those becomes a property of the record itself: signed, witnessed, and independently verifiable, at the moment the decision is made.
How it works
Step 01
VAiDR sits where your agents act: an inline control plane over every model and tool call. It captures the metadata of each action: which agent, which human authorized it, which policy version was in force, which model ran. Never your prompts or outputs.
Step 02
You author the policy, not us. It runs in the request path, before the model is called. Block the action, pause it for a human, allow it with a flag, or allow it. High-stakes rules fail closed: if the check can't run, the request doesn't either.
Step 03
Every decision becomes a signed record, hybrid Ed25519 + ML-DSA-65, witnessed in a public transparency log within seconds. Anyone with the record and your public key can prove what happened, years later, without trusting Stones AI or the operator.
Enforcement
Monitoring observes; it never intervenes. By the time an alert fires, the data has left, the claim is denied, the funds have moved. For agents acting at machine speed, the only control that matters happens before the action, and every enforcement decision is recorded and signed.
The request is rejected before the model is ever contacted. A signed "blocked" record is still emitted.
The request is held for operator approval. A human approves or rejects before anything proceeds.
The request proceeds, but the record is marked with the policy match for audit and review.
The request proceeds normally and a standard signed record is written.
You author the policy. We don't decide what's safe for you. Rules are plain, auditor-readable conditions on prompts, tools, models, agents, and thresholds. Each rule declares whether it fails closed or fails open when a check can't run, and that choice is itself part of the auditable policy.
Data, privacy & security
One rule: your confidential data stays in your environment. Stones AI never sees it, never stores it, never holds your keys. What crosses the line is proof a decision happened, with none of the content.
Inside your VPC · stays with you
Leaves your VPC · what we witness
The signing key is generated and held by you, never by Stones AI. Records are witnessed in the public Sigstore Rekor transparency log plus an independent witnessed log. Disagreement between the two is itself the tamper signal. Anyone can verify the record. No one can read your data.
Who it's for
Financial services, insurance, legal, healthcare: anywhere AI actions have to be answered for. Regulated industries are converging on one demand: a durable, inspectable record of what your AI did.
Your examiners are converging on one demand: a durable, inspectable record of what your AI did, provably unaltered. VAiDR produces exactly that record at runtime, in the same format for the regulator, the court, and the insurer.
Guardrails today are vendor-decided and after-the-fact. VAiDR puts your rules in the request path: enforcement before the action, not an alert after the data has left.
You're liable for the AI's action either way. The question is whether you can prove what it actually was. A plain-text log is weak evidence; a signed, independently witnessed record is a chain of custody.
Why us
Trust is the product
The deliverable is evidence: a record a regulator, a court, and an insurer can all trust, without trusting us. VAiDR doesn't ask anyone to take your AI's behavior on faith, including you.
Read the principleHumans always decide
You author the policy. High-stakes actions pause for a human before anything proceeds. The record binds every AI decision to the person responsible for it. Accountability with a name on it.
Read the principleShow your work
Every decision is signed, publicly witnessed, and verifiable by anyone with the record and your public key, years later, no matter what happens to the vendor. That's showing your work, cryptographically.
Read the principleWhere we are
VAiDR is built and tested end-to-end and in production hardening. Design-partner pilots start in H2 2026.
Shipped
Signs a record at the point your agent acts.
Shipped
Sits inline over HTTP and enforces your policy on every call.
Shipped
See and verify every record VAiDR creates.
In build
Identity for every agent that acts.
Design partners
If you're deploying AI agents in a regulated environment and can't yet answer for what they did, we want to talk. Design-partner pilots start in H2 2026.
