LLM Providers

GCTRL runs on local inference by default and can optionally connect to cloud providers when you want them. You choose the trade-off between performance, cost, and data sovereignty.

Local by default: Ollama

Out of the box, GCTRL uses local Ollama:

• On-prem. Inference runs on your hardware.
• Zero token cost. No per-request billing.
• Data-sovereign. No prompt or content ever leaves the machine.

This is the recommended path for regulated and GDPR-sensitive deployments.

Optional cloud providers

For teams that want frontier models or higher throughput, GCTRL also supports:

Configuring a provider

Connect providers in Settings → connect provider:

• Cloud keys are encrypted at rest and are never returned by the API - once saved, a key cannot be read back out.
• Cloud base URLs are pinned and SSRF-guarded - GCTRL only talks to the provider's known endpoint, preventing request forgery to internal hosts.

Embeddings

Embeddings default to nomic-embed-text, run locally. This keeps the vector index - and the content it represents - on your machine regardless of which generation provider you select.

Choosing for data sovereignty

For full data sovereignty, use Ollama only. Any cloud provider you connect will, by definition, send prompts to that provider. Leave cloud providers disconnected to guarantee nothing leaves your infrastructure.

See also

• Infrastructure - switch to native GPU-accelerated Ollama for a major speedup
• Performance - pick the right model for your hardware