Your data
Short answer: on Solo with local models, essentially nothing. Full answer: below, with the diagram, the mechanisms, and the limits.
Processed on your device. Full stop.
Always your choice, never a hidden default.
Inference can run entirely on your device (llama.cpp, GGUF models). So can transcription (Parakeet ONNX). If you plug in your own API key, requests go straight to your provider under your own agreement — Luge is not a middleman reading your traffic.
Your secrets (keys, integration passwords) resolve locally through 1Password. The session token lives in the operating system's keychain, not in a config file.
PI detection (SIN, health-insurance numbers, emails, phone numbers, cards) runs locally inside the platform — content is not shipped to an external service for analysis.
Every significant event lands in audit logs sealed at write time (a per-row integrity hash — tampering is detectable by a third party), exportable as PDF evidence packages. Every resource is isolated per organization. That's the architecture our first vertical customer — Canadian financial services — required, applied to everyone. The illustrated version is here.
Need managed inference to stay in the country? The managed sovereign modules run on Canadian infrastructure managed by us — the supplier can evolve, the residency doesn't. Details on the pricing page.
This site uses no tracking cookies and needs no banner. Visit statistics are aggregated and anonymous.
Luge is designed to make your compliance easier: local processing, personal-information detection, audit logs, and an option to run inference on infrastructure hosted in Canada. Compliance with Law 25 remains an organizational responsibility — no software can "guarantee" it. If a vendor promises otherwise, worry about the vendor.
Luge's first vertical deployment is operated by Tchat N Sign, whose organization holds SOC 2 Type II, ISO 27001 and ISO 27701 certifications.
Our full privacy policy is here. Questions? [email protected].