Will Intellirity slow down our AI?

No. Verdicts compute inline with async enforcement, so users never wait on security. The live round-trip is measured on every scan — open the dashboard and watch the mesh-health readout in the sidebar.

Which models and frameworks are supported?

Anything that speaks text: OpenAI, Anthropic, Google, Mistral, Llama, plus LangChain, Bedrock, Vertex, and raw HTTP. Register each endpoint under Models & agents and scan it against the live engine.

Can we attack it ourselves before buying?

Please do. Paste a jailbreak into the live scanner and watch the verdict stream, or run a sweep across all sixteen detection engines — see the full engine list.

Where does our data live?

With you. Verdict metadata is yours and we never train on traffic. Every decision lands in the tamper-evident evidence vault, verifiable any time. A sovereign deployment runs the whole engine inside your own VPC.

What happens when the AI tries something risky?

Dangerous actions are held in agent escrow until a human approves them, and every approval carries a signed proof of intent.

How is this different from a WAF?

WAFs understand HTTP. Intellirity understands intent: semantic injection, multi-turn jailbreak chains, agent abuse, and exfiltration. See each control mapped to its live tab on the platform section.

How much does it cost?

Intellirity is billed per detection module, per month: nine modules from ₹79 to ₹399. There are no tiers and no seat counts — you toggle the coverage you need and pay the sum. Taking all nine together is 20% off the total. Compare every module in pricing.

Are you SOC 2 or ISO 27001 certified?

Not yet. We have not completed an independent SOC 2 Type II or ISO/IEC 27001 audit, and we do not publish badges we cannot evidence. What the product does enforce is written down in the security model.

AI Security Model

Effective 1 January 2026

How the Intellirity AI security model works: fail-closed screening, tamper-evident audit evidence, tenant isolation, no traffic retention, and coordinated vulnerability disclosure.

Screening fails closed

If a detection engine errors, the screening proxy denies the request rather than forwarding it. A crash must never become a bypass.

An invalid API key is rejected with 401 even when the deployment is in open mode, and a session token presented to a deployment that cannot validate it returns 503 instead of silently granting anonymous access.

Evidence is signed and chained

Every ledger record is sealed into a Merkle-style hash chain: each record commits to the one before it, so a removed or altered record breaks the chain and verify reports it.

Proof-of-intent certificates are signed with HMAC-SHA256 using a server-side key. A certificate that was not issued by this server is rejected, as is one that has expired or whose proposed action falls outside its signed scope.

Tenant isolation

Ownership is scoped to a tenant: a member of a Clerk organization reads and writes that organization's billing, model endpoints, policy packs and escrow holds. Without an organization, a user owns a private tenant. Machine API keys are their own tenant and are never written to the user table.

Engine responses are served with `Cache-Control: private, no-store` and `Vary: Authorization`, so a shared cache can never replay one tenant's data to another.

Input handling

Prompt payloads are processed to produce a verdict and are not retained by default.

Code scanning enforces a size cap and refuses to fetch private, loopback, link-local or reserved addresses, so a scan target cannot be used to reach internal services.

The console and public API are served with a Content-Security-Policy, HSTS, strict referrer policy and denied framing.

Deployment options

Managed: the console and API run as a hosted deployment with Postgres for durable storage.

Sovereign: the engine ships as a sidecar container you run inside your own VPC, so prompts never leave your network.

Reporting a vulnerability

If you find a vulnerability in Intellirity, report it to security@intellirity.com. We acknowledge within 24 hours, ship a fix or mitigation within 30 days, and credit researchers publicly unless you prefer anonymity. Good-faith research is authorized and will not be pursued legally.

Intellirity has not completed an independent SOC 2 Type II or ISO/IEC 27001 audit. We do not claim certifications we cannot show you. If those matter to your procurement process, ask us for the current control documentation.