CirvixAI agent security

AI agent security you can inspect.

Q: What is AI agent security?

AI agent security is the control layer between an autonomous agent and everything it can touch: files, tools, APIs, data. Cirvix checks each proposed action against policy before it runs and records the decision in the unsigned hash-chained log.

Q: Can I try it without an account?

Yes. npx @cirvix_ai/agent-control scan runs the enforcement engine locally, sends nothing anywhere, and needs no signup. The free tier is $0 forever.

Agents act: they read files, call tools, move data. Cirvix is the AI agent security layer that checks every one of those actions against policy before it runs — built on engineering mechanisms you can inspect instead of unverified marketing claims. Where something is in progress, this page says so. For the full category guide, see AI-agent security.

See the architecture →Request a review

AES-256-GCMenvelope encryption

SHA-256audit chain linking

0unsanctioned retention

In progressSOC2 / ISO certification

Verifiable, not assumed

The mechanisms.

Each of these is a structural property of the system rather than an operational promise.

✓

AES-256-GCM envelope encryption

CIRVIX_MASTER_KEY is customer managed. No escrowed master key path.

✓

Tamper-evident audit chain

Cryptographically linked SHA-256 logs preserve the integrity of each execution.

✓

Structural tenant isolation

org_id row-scoping is enforced in routing rather than added as an application convention.

✓

Zero unsanctioned retention

Decisions enter the audit chain; payload data is discarded when it is not approved for retention.

✓

OIDC / SCIM support

Native integration points for enterprise identity providers and managed access.

✓

Automated evidence generation

Export a coherent audit trail for a review without assembling raw logs by hand.

Cirvix enforces strict structural security mechanisms. Formal SOC2 and ISO certifications are currently in progress; we publish no unverified SLA claims.

Identity

Who an agent
actually is.

Identity mistakes are the cheapest way to lose a tenant boundary, so identity is resolved structurally.

Identity is (issuer, subject)

Never email. An email address can be reassigned inside a directory; an issuer-plus-subject pair cannot be silently inherited by someone else.

SCIM is a separate credential surface

The SCIM token is not a console key and cannot be used as one. This is enforced by test, not by documentation.

Deprovisioning revokes sessions

Removing a user ends live sessions rather than only removing membership, so an open session cannot outlive the account.

SAML is deliberately not supported

OIDC is implemented for Google, Entra, and Okta. SAML was refused rather than half-built — if you need it, that is a real gap and we will say so.

Data retention

What we keep,
and for how long.

The audit chain is a record of decisions. It is not a copy of your data.

retention policyDEFAULT

decision records: Retained in the audit chain for the life of the deployment
rule + reason: Retained — this is what makes a record explainable
payload contents: Discarded unless explicitly approved for retention
master key: Never held by Cirvix. Customer managed, no escrow
self-hosted: Retention is bounded by your own store, not ours

Posture

Where we are
being honest.

Five things are worth stating plainly rather than burying in a trust page.

Item	Status	Note
SOC 2	In progress	Not certified today. We do not describe ourselves as compliant.
ISO 27001	In progress	Evidence reports are generated; certification is a separate process.
Uptime SLA	Not published	We publish no unverified SLA claims. Contractual terms are agreed per deployment.
Evidence reports	Available	Coverage reports never claim compliance — the vocabulary has no word for “pass”.
Company structure	Sole proprietor	Cirvix is operated by Umang Kumar, trading as Cirvix (see Terms). Weigh it as you would any early-stage vendor: there is no established enterprise behind this product.

Item

Status

Note

SOC 2

In progress

Not certified today. We do not describe ourselves as compliant.

ISO 27001

In progress

Evidence reports are generated; certification is a separate process.

Uptime SLA

Not published

We publish no unverified SLA claims. Contractual terms are agreed per deployment.

Evidence reports

Available

Coverage reports never claim compliance — the vocabulary has no word for “pass”.

Company structure

Sole proprietor

Cirvix is operated by Umang Kumar, trading as Cirvix (see Terms). Weigh it as you would any early-stage vendor: there is no established enterprise behind this product.

Responsible disclosure

Found something? Report it and we will respond with the actual state of the fix, not a receipt.

Report an issue →

FAQ

AI agent security,
asked directly.

Short answers with no marketing padding. Each one matches the behavior enforced in the engine.

What is AI agent security?

It is the control layer between an autonomous agent and everything it can touch: files, tools, APIs, data. Cirvix checks each proposed action against policy before it runs and records the decision in the unsigned hash-chained log, so nothing executes unchecked.

How is this different from prompt-injection detection?

Detection flags bad text after the fact. Cirvix enforces a boundary before the action: even if an agent is tricked by injected content, the downstream tool call it attempts is still evaluated and can be denied. The block holds regardless of what fooled the model.

Does Cirvix slow agents down?

Policy evaluation is designed against a millisecond budget per stage, and refused calls are not counted against the free daily quota. The engine runs local-first, so there is no round trip to a hosted service on the decision path.

Where does my data go?

Nowhere unsanctioned. Approved decision evidence enters the tamper-evident audit chain; payload data that is not approved for retention is discarded. Self-hosted deployments keep everything inside your own boundary.

Can I try it without an account?

Yes. npx @cirvix_ai/agent-control scan runs the enforcement engine locally, sends nothing anywhere, and needs no signup. The free tier is $0 forever with 100 decisions per day.

Govern what ships

Bring every agent
under control.

Set durable policy, preserve a verifiable record, and give teams a safer way to put intelligent systems to work.

Launch Control Plane →Read the docs