Skip to main content
SecuAI capabilities

Sovereign AI for your sensitive workloads

Chat, embeddings, completions, specialized models — hosted at OVH Beauharnois, with no training on your data.

Sovereign LLMs

Qwen3 35B-A3B (sovereign) and Mistral Small 24B, FR/EN specialized models. Sovereign model on AMD R9700 GPU in Canada.

PII anonymization

Automatic detection of names, emails, Canadian SIN, RAMQ HCN, addresses, IBAN, credit cards. Reversible tokenization.

Integrated OVH KMS

Encryption of API keys and payloads via OVH KMS Beauharnois. Hardware-bound HSM available. Immutable audit trail.

OpenAI-compatible API

Drop-in replacement: just change the base_url. Compatible with LangChain, LlamaIndex, Haystack, Semantic Kernel, OpenAI SDK.

Embeddings & RAG

Multilingual embeddings, tenant-aware Haystack integration, pgvector, Qdrant, Weaviate. RAG over internal documents.

BYOE (Bring Your Own Encryption)

Client-side payload encryption before transit, decryption only inside HSM, never plaintext on SecuAI side.

MSP multi-tenant

Partner console, per-tenant isolation, quotas (req/min, tokens/day), white-label branding, distributor billing.

Loi 25 compliance

Quebec/Canada PII natively detected, pre-built registries, Loi 25 article mapping, immutable signed receipts.

Store-original opt-in

By default, no storage of the original prompt. Explicit per-tenant opt-in for debug/audit, configurable 24h-30d TTL.

No third-party transit

No calls to OpenAI/Anthropic/Google. 100% inference at OVH Beauharnois. No data ever trains an external model.

Quebec residency

Inference, storage, embeddings and logs are 100% at OVH Beauharnois — outside Cloud Act/Patriot Act/FISA.

Python/Go/Node SDK

Official SDKs with BYOE helpers, exponential retry, SSE streaming, complete type hints, OpenAI compatibility.