FoundiX

VOLT Lattice · adjacent, not a SovSCC SKU

Sovereign inference, when the weights are ours.

Phase-1 public answers are extractive from a pinned corpus. Every answer carries a SHA-256. Phase-2 attaches FoundiX-owned open weights behind the same URL. Nothing here is OpenAI, STACKIT Model Serving, Uniserver, or Sitejet.

  1. 01

    Phase 1 — now

    Extractive answers from a hash-pinned corpus. If the corpus does not contain it, we refuse. No vendor LLM.

  2. 02

    Phase 2 — HOLD

    FoundiX-owned open weights on owned metal (Lattice M-INFER). Same public URL. GPU overflow is never the default.

  3. 03

    Phase 3

    Train and finetune on tenant datasets that do not leave the perimeter.

Rejected as default

  • OpenAI API
  • STACKIT Model Serving
  • Uniserver ChatGPT interface
  • Sitejet / ai.sitehub.io
  • Hyperscaler Copilot control planes