// project · /catalog · experimental · type=service

OnPremAI

Governed AI access layer. Two-plane airlock keeps prompts inside the perimeter — internal plane runs scoped knowledge sources, external plane handles general research, sanitization + approval boundary between them.

entity

────────────────────────────────────────────────────────────────

apiVersion: backstage.io/v1alpha1
kind:       Component
metadata:
  name:        onpremai
  title:       OnPremAI
  tags:        [cloudflare-pages, workers-ai, kv, d1, r2, turnstile, airlock, governed-ai, opa, rego]
spec:
  type:        service
  lifecycle:   experimental
  owner:       anil-ardic
  system:      anilardic-portfolio
  providesApis: [airlock-contract, policy-engine-rego, governed-agents]

annotations

────────────────────────────────────────────────────────────────

anilardic.com/repo: /home/anil/projects/onpremai
anilardic.com/design-repo: /home/anil/projects/onpremai/onPremAI
anilardic.com/demo-repo: /home/anil/projects/onpremai/onpremai-demo

links

────────────────────────────────────────────────────────────────

CANONICAL Demo (CF Pages + Workers AI) https://onpremai-demo.pages.dev

notes

────────────────────────────────────────────────────────────────

Design phase + working demo: airlock-contract JSON schema, OPA Rego policy engine, redaction corpus, threat model (STRIDE T1–T12), 4-agent governance core (M1–M6 in-scope; support + document deferred). Demo proves the governance UX is real: paste a synthetic incident, watch the 4-stage airlock, approve, get a signed audit-log entry. Real product runs on-prem.

← back to /catalog · /