Technology

Sovereign & On-Prem

GCTRL runs entirely on hardware you control - data never leaves by default, cost is deterministic, and the deployment can go fully air-gapped.

What it does

Every component - the graph store, the vector index, the relational store, the job queue, and inference itself via local Ollama - runs on infrastructure you operate. Nothing about a document, a query, or an answer is required to touch a third party's servers. Because the extraction and fusion pipeline runs locally rather than paying per token to a cloud model, cost per document is deterministic - it doesn't scale with an external API's pricing, and it doesn't stop working if a provider changes terms or goes down.

Why it matters / USP

For regulated industries and any organization that has been told "you can't put that in the cloud," on-prem isn't a nice-to-have, it's the entry requirement. GCTRL is built so that requirement doesn't cost you frontier-model-quality extraction and retrieval: the entity detection and entity-resolution numbers on the Benchmarks page are measured running fully locally, with zero training data and zero cloud dependency. Because nothing is architecturally required to leave the network, the deployment can go as far as air-gapped - no outbound connectivity at all - for the environments that demand it, while an organization comfortable with selective cloud use can still opt in per graph via Cloaking.

Sovereignty here is also a cost argument, not just a compliance one: a per-token cloud bill scales with usage in a way that's hard to forecast at enterprise volume; a deterministic, on-prem per-document cost doesn't.

How it fits

Sovereignty is the deployment posture underneath the whole pipeline - KEX, FUSE, Manage KGs, and Talk-to-Graph all run locally by default, and every backing store (graph, vectors, relational, queue) is yours to operate and swap for your own managed equivalent.

In practice

A defense contractor needs an air-gapped deployment with no outbound network access whatsoever. GCTRL runs the full pipeline - extraction, fusion, retrieval - against local Ollama and local stores with connectivity to nothing outside the perimeter, and the classification and audit trail work exactly the same as they would in a cloud-connected install.

See also

Compliance & Data Sovereignty · Cloaking · Classification & Access Control