Skip to main content
For Design Partners

Platform Architecture

A security-first identity and access management layer purpose-built for autonomous AI agents. Every request is authenticated, authorized, and audited before it reaches an LLM provider.

System Overview

Every AI agent request flows through the AI Identity Gateway before reaching any LLM provider. The gateway enforces policy, manages credentials, and writes a tamper-evident, OCSF-formatted audit record — plus a signed attestation anyone can verify offline, with nothing but a public key.

┌──────────────────┐ │ Your AI Agents │ │ (SDK / API) │ └────────┬─────────┘ ▼ ┌──────────────────┐ │ AI Identity │ │ Gateway (Proxy) │ └────────┬─────────┘ ▼ ┌──────────────────┐ │ Policy Engine │ │ (Real-time) │ └────────┬─────────┘ ▼ ┌──────────────────┐ │ Credential │ │ Vault │ └────────┬─────────┘ ▼ ┌──────────────────┐ │ LLM Providers │ └──────────────────┘ ┌──────────────────┐ │ Audit Trail │ │ (OCSF · chain) │ └────────┬─────────┘ ▼ ┌──────────────────┐ │ Verifiable │ │ Evidence · │ │ Case File │ └──────────────────┘ ┌──────────────────┐ │ Dashboard │ └──────────────────┘
Request flow Encrypted at rest & in transit Signed OCSF evidence, verifiable offline
See a verifiable Case File

Key Architectural Properties

The design decisions that make AI Identity enterprise-grade, compliance-sensitive, and forensically defensible — every agent action provable after the fact.

Fail-Closed Enforcement

Any error in the policy evaluation pipeline results in an automatic deny. Agents cannot bypass controls, even during partial outages.

Real-Time Policy Enforcement

Policy decisions are evaluated inline with less than 50ms of added latency. No queued evaluation, no eventual consistency — enforcement happens before the request proceeds.

Credential Vault

LLM provider API keys are stored in an encrypted vault. Agents never see or handle raw credentials — the gateway injects them at request time.

Zero-Trust Key Separation

Runtime keys used by agents are fully separated from administrative keys. Compromising one does not compromise the other.

Tenant Isolation

Row-level security ensures each tenant's data is completely isolated. One customer's agents, policies, and audit logs can never leak into another's.

Tamper-Evident Audit Chain

Every event is cryptographically chained with HMAC-SHA256. Alter any record and the chain breaks — making tampering detectable and provable.

OCSF-Native Records

Every audit event is emitted in the Open Cybersecurity Schema Framework — the open standard we contribute to upstream. Records drop straight into Splunk, your SIEM, or any OCSF tooling, no proprietary format to reverse-engineer.

Independently Verifiable Evidence

Each forensic export carries an ECDSA-signed DSSE attestation. Auditors verify it completely offline with nothing but a public key — no access to our systems, no trust in the vendor required.

Portable Case File Export

Package any incident into a self-contained Case File — signed OCSF records plus a drag-and-drop verifier. Hand it to counsel, an auditor, or a regulator and they can confirm integrity themselves.

Security Layers

Defense in depth — multiple independent layers ensure no single failure compromises the system, from the first byte on the wire to the evidence an auditor verifies months later.

TLS Everywhere

All traffic is encrypted in transit using TLS 1.2+. No plaintext communication between any components.

JWT Authentication

Every API request is authenticated via signed JSON Web Tokens with short-lived expiration and audience validation.

Rate Limiting

Configurable per-agent and per-tenant rate limits prevent abuse and protect downstream LLM provider quotas.

Policy Enforcement

Fine-grained rules control which agents can access which models, with what parameters, and under what conditions.

Encrypted Credential Storage

All sensitive credentials are encrypted at rest using AES-256. Decryption only occurs in-memory at request time.

OCSF-Native Audit Events

Every action is recorded as an Open Cybersecurity Schema Framework event — a vendor-neutral standard your SIEM already understands, no proprietary log format to parse.

Tamper-Evident Audit Chain

Audit events are sequentially chained with HMAC-SHA256. Any modification to a historical record breaks the chain and is cryptographically detectable.

Signed Forensic Attestation

Forensic exports are sealed in an ECDSA-signed DSSE envelope. Anyone can verify integrity and provenance offline with a public key — no access to AI Identity required.

Compliance Alignment

Built from the ground up to satisfy the requirements your security and compliance teams care about most.

SOC 2

Architectural principles aligned with SOC 2 Trust Services Criteria.

  • -Logical access controls with key separation
  • -Tamper-evident, independently verifiable audit trail
  • -Encrypted data at rest and in transit
  • -Tenant isolation with row-level security

NIST AI RMF

Designed to support the NIST AI Risk Management Framework principles.

  • -Complete observability of AI agent actions
  • -Policy-based governance and enforcement
  • -Signed, verifiable evidence for accountability
  • -Fail-closed design for reliability

EU AI Act Ready

Infrastructure controls that support EU AI Act obligations for high-risk AI systems.

  • -Human-in-the-loop policy overrides
  • -OCSF audit log for end-to-end traceability
  • -Transparent enforcement decisions
  • -Data residency-aware architecture

Ready to see it in action?

Walk through a live demo, or verify a signed Case File yourself — no account, no trust required.

Live DemoExplore AI ForensicsSchedule a Call