The platform

A single Helm chart. Your entire AI estate, governed.

Perimeter deploys as a set of Kubernetes microservices into your existing cluster. No agents to install on every workstation, no SaaS to subscribe to, no metadata leaving your network.

AI Discovery Engine

  • Cluster-wide GPU workload scanning via DCGM, ROCm-SMI, XPU Manager.
  • Auto-detection of vLLM, TGI, Ollama, Triton, and custom model servers.
  • Shadow-AI flagging for unregistered model deployments.

Risk Classifier

  • Auto-mapping to EU AI Act risk tiers (prohibited / high / limited / minimal).
  • NIST AI RMF function alignment (Govern, Map, Measure, Manage).
  • ISO 42001 control evidence generation.

Runtime Monitor

  • Envoy sidecar or standalone gateway. <50ms P99 for inline checks.
  • Local NER models for PII, locally-run bias and fairness evaluators.
  • Hallucination scoring via vLLM/TGI logprobs and RAG cross-reference.

Model Supply Chain

  • Provenance tracking from Hugging Face, internal registries and fine-tunes.
  • License compliance checks (Apache 2.0, Llama community, restricted-use).
  • Data poisoning scans, backdoor heuristics, weight diffing.

Compliance Automation

  • Continuous technical documentation generation for high-risk systems.
  • Conformity assessment workflows with sign-off trails.
  • Post-market monitoring records, drift triggers and re-evaluation jobs.

Audit & Integration

  • Append-only audit ledger. Immutable. Exportable.
  • Native syslog/webhook out to Splunk, Elastic, ServiceNow GRC, Archer.
  • OPA-backed RBAC, mTLS via cert-manager, encryption at rest.
Stack

Built on the infrastructure you already run.

Backend
Go (control plane, scanners), Python (ML evaluators)
Runtime
Kubernetes 1.27+, OpenShift, Rancher, vanilla K8s
Storage
PostgreSQL (operator-managed), MinIO object storage
Security
mTLS via cert-manager, OPA + K8s RBAC, LUKS at rest
Observability
Prometheus, OpenTelemetry, Grafana dashboards
Deployment
Helm chart, Operator, signed OCI artifacts for offline updates