AI security

Trustworthy in the agent's critical path.

Meandr sits between your agents and the tools they call, so every request runs through it. That only works if it's built to be trusted — encrypted end to end, blind to your payloads by default, and honest about what it does and doesn't hold.

Get started How capture stays blind

Encryption everywhere.

Traffic is encrypted on the way in, on the way out, and anywhere it comes to rest. There is no plaintext leg in the path.

TLS in, with SNI required

Agents reach Meandr over TLS, and every handshake must name the host it's for. A connection that arrives without SNI — an IP scanner, a raw-port probe — is refused at the handshake, before it reaches anything.

mTLS to your upstreams

For servers that require a client certificate, mTLS is a first-class upstream auth type. Meandr presents the client cert on the handshake, so a server can be locked to callers it can cryptographically verify.

Upstream TLS verified by default

Meandr re-encrypts every call to the real server and verifies its certificate. The only way to turn that off is an explicit per-server flag — and choosing it writes an audit signal, so an insecure upstream is never silent.

Encrypted at rest, decrypt-on-view

Anything Meandr is asked to keep — a captured payload — is encrypted at rest under a KMS key. The proxy only ever encrypts; the plaintext is decrypted solely inside an authorized "show this request" read, never at rest and never in bulk.

Payload-blind by design

We can't read your traffic — even when you ask us to keep it.

By default Meandr logs metadata about a call, not the call's contents. Keeping the bodies is a choice you make per policy — and even then, keeping is all it does.

Metadata by default

Every call leaves a small, immutable record — which tool, which agent, the policy decision, status and timing — on the order of a couple hundred bytes. The request and response bodies are not part of it.

Capture is opt-in, per policy

A policy can choose to store the request, the response, both, or neither. So "capture everything except payroll" is a rule you can actually write — and a metadata-only policy genuinely stores nothing.

No DLP, no regex, no PII scanning

Capture is storage, not inspection. Meandr never parses a stored body looking for secrets or personal data — that would break the blindness this is built on. What you keep is yours to read, not ours.

How the platform is built.

The security story isn't only the traffic. It's how the platform behind it holds credentials, stores data, and keeps standing access to a minimum.

Near-zero standing credentials

The platform is built to run without long-lived keys sitting around — access comes from short-lived, role-based identities rather than static secrets, so there is very little for a leak to hand over.

Private, encrypted data stores

The stores behind Meandr are private and encrypted at rest, reached only over verified, mutually-authenticated connections. There is no public door and no plaintext hop between the moving parts.

Your upstream keys, encrypted

The credentials Meandr uses to reach your servers — API keys, OAuth tokens, cert bundles — are sealed with AES-256-GCM under a KMS-wrapped data key, one dedicated key per environment, and decrypted only in memory at the moment a call needs them.

Compliance

Built to SOC 2 — certification is the next step.

Meandr was built to SOC 2 from day one — the controls are in place, not bolted on later, and we're preparing for the audit. Obtaining the certification itself is a deliberate next step after launch. We'd rather say exactly where we stand than imply a certificate we don't yet hold.

Ask about our security posture

HIPAA is a design goal too — the custody model is being built to stand up to that scrutiny — but it's a direction, not a claim of compliance.

Questions you'd ask anyway.

Does Meandr store the contents of my tool calls?
Not by default. By default it keeps only metadata about each call — tool, agent, decision, status, timing — never the bodies. Storing request or response payloads is something you turn on per policy, and even then it's stored, not inspected.
If I do capture payloads, who can read them?
Captured bodies are encrypted at rest under a KMS key. Meandr's proxy only ever writes them encrypted; they're decrypted only inside an authorized "show this request or response" view, never at rest and never in bulk. There is no scanning of the contents.
How are my upstream credentials protected?
Your servers' keys and tokens are sealed with AES-256-GCM under a KMS-wrapped data key — one dedicated key per environment, separate from anything else and never kept in the config cache. They're decrypted in memory only when a call to that server needs signing, and your agents never see them.
Are you SOC 2 certified?
Not yet — and we won't say so until we are. The controls are in place and we're preparing for the audit; running it is a planned step after launch. If you need to know where a specific control stands, ask us and we'll tell you straight.

Put an agent's traffic somewhere you can trust.

Point your agent at one endpoint and you're governed in minutes — no code changes, no proxy to run.

Get started

Questions? Email us