Product

Sovereign AI Platform

A privacy-first AI platform designed to run fully on customer-controlled infrastructure (including air‑gapped environments) for teams that cannot send sensitive data to cloud AI APIs.

Core capabilities (Alpha 1 direction)

  • Local inference for key workflows (text + text→image)
  • Model lifecycle: import / pin / rollback
  • Security baseline: RBAC + audit logs (GDPR-first)
  • Air‑gap operations: offline updates + exportable diagnostics

Initial workflows

  • Document summarization & extraction (sensitive corpora)
  • Internal ops/support chatbots
  • Text→image “infographic” generation (poster-style)

Out of scope for Alpha 1: fine-tuning/training, outbound allowlists (web search), enterprise SSO.

Benchmark-first

We make “performance vs OpenAI” reproducible instead of hand-wavy: a fixed eval harness with domain quality (X) and user-capacity throughput (Y) targets.

  • X: ≥ 80% pass rate on domain eval suite (healthcare, finance, IP-heavy SMEs)
  • Y: N=3 concurrent users, p95 ≤ 30s (text summary) and ≤ 120s (text→image)
  • Baseline for text/domain evals: OpenAI gpt-4o-mini on non-sensitive eval inputs