SecuAI: generative AI with sovereign anonymization, in Québec
SecuAI is a generative AI gateway whose sovereignty rests on anonymization. Enabled on a request or enforced on a key, it detects personal information entirely in Québec: the original information never leaves Québec to be analyzed. Each item is replaced with a consistent pseudonym, or the request is blocked, depending on the mode chosen.
SecuAI
Get the benefits of generative AI. Keep your personal information in Québec.
from $29 / month
| Anonymization (SecuAI) | Included |
|---|---|
| Anthropic (Claude) | $0 / month — usage billed |
| DeepSeek | $0 / month — usage billed |
| Google (Gemini) | $0 / month — usage billed |
| Sovereign GPU (Quebec) | $0 / month — usage billed |
| Mistral (La Plateforme) | $0 / month — usage billed |
| OVH AI Endpoints (EU) | $0 / month — usage billed |
| Telnyx (GLM-5.2, United States) | $0 / month — usage billed |
| xAI (Grok) | $0 / month — usage billed |
| Z.ai (GLM) | $0 / month — usage billed |
Why SecuAI
Your teams already use AI.
The question is no longer “if”, but “with what protections”.
Quebec's Law 25 also applies to the prompts you type.
A name, an email, a file number in a question to a foreign model is a disclosure of personal information outside Québec.
A single control point.
Every AI call in your organization goes through SecuAI: anonymization, traceability, cost centralization, audit.
You choose the model.
Sovereign or third-party, with the country of residency displayed for each.
Two ways to use it
Anonymization only
You keep your current AI tools. You call SecuAI through the API to anonymize your texts, documents or streams before sending them wherever you like, then to re-identify them on return. You control the chain end to end.
AI connectors with built-in anonymization
You go through SecuAI connectors to the major market models (OpenAI, Anthropic, GLM and others). Anonymization is enabled by a simple header on each request, or the SecuAI team can enforce it on a key: your users then have nothing to do. A slight surcharge covers anonymization, traceability, centralization and audit.
What you get
Four-layer anonymization
Safe formats, text understanding, a second look by AI and a final review, then an anti-leak check before sending. Two modes: tokenization or blocking. Details below.
Sovereign models
Models hosted in Québec, on GPUs in Québec.
Third-party models, residency displayed
Each third-party model is shown with its country of residency (United States, European Union, China and others), so you choose with full knowledge.
Traceability and audit
Who called which model, when, at what cost, with what level of anonymization. One centralized log for the whole organization.
Prepaid AI credits
Buy credits, distribute them among your teams, track consumption from the console.
100% API
Compatible with your applications, your automations and your agents.
SecuAI anonymization: four layers, one promise
Before a request leaves for an AI model, SecuAI can remove the personal information it contains: names, emails, phone numbers, SINs, cards, addresses… Each item is replaced with a consistent pseudonym. The response comes back to you with the real values restored. Detection takes place entirely in Québec, on SecuAAS infrastructure: the original personal information never leaves Québec to be analyzed.
Two modes to choose from for sensitive information
- Tokenization: each item is replaced with a consistent pseudonym, and the response comes back with the real values restored.
- Blocking: the request is refused at the first secret detected.
Four layers of detection
- L1, safe formats: precise rules recognize identifiers with a fixed structure and verify their checksums: SINs, credit cards, IBANs, emails, phone numbers. This layer also defeats common disguises: invisible characters, look-alike letters, digits from other alphabets, encoded content.
- L2, text understanding: entity recognition models, specialized in personal information, spot what no rule catches: a proper name, an organization, a freely written address.
- L3, a second look by AI: a language model hosted on SecuAAS GPUs, in Québec, rereads the text to find what the first two layers missed.
- L4, the final review: a corrector, also sovereign, compares the original text with the anonymized text. It can catch an overlooked item or restore a word masked in error. It never restores a password or a login identifier.
The safety net
Just before sending, a final check verifies that no original information survived. If any did, the request is blocked. Likewise, if L2 detection is unavailable, the request is refused rather than sent with less protection.
What you control
| Setting | You (client) | SecuAI team |
|---|---|---|
| Enable anonymization on a request | Yes, with a simple header | Can enforce it on a key |
| Choose the categories to mask | Yes (22 by default; amounts and places optional) | Yes |
| Restore the real values in the response | Yes, as an option | — |
| Disable layers | L2 and L3, at your own responsibility; L1 and L4 always active | L2 and L3 by configuration; L4 is never turned off |
| Disable the final anti-leak check | No | Internal configuration only |
- Disabling L2 or L3 reduces protection: this decision is the client’s responsibility.
- L1, L4 and the final anti-leak check remain active when anonymization is enabled.
- If you disable L2, its unavailability does not block the request: the “L2 unavailable” refusal applies only when L2 is enabled.
Sovereignty
- Detection of personal information entirely in Québec, on SecuAAS infrastructure: the original information never leaves Québec to be analyzed.
- Two modes to choose from: tokenization with consistent pseudonyms, or blocking at the first secret detected.
- Final anti-leak check before every send: a request in which original information survived is blocked.
- Sovereign models on GPUs in Québec; residency of each third-party model displayed.
- Designed for Quebec's Law 25.
How it starts
- Create your Olimar account.
- Activate SecuAI and choose your mode: anonymization only or connectors with built-in anonymization.
- Buy your AI credits; use SecuAI from the console or through the API.
- Olimar can integrate SecuAI into your applications, define your anonymization policies and train your teams, as scoped.
Who it's for
Professional firms (legal, accounting, real estate), health and social services, public bodies, SMBs that want to govern the use of AI without slowing it down, developers who integrate models into their products.
Frequently asked questions
Is anonymization on by default?
No. You enable it on a request with a simple header, or the SecuAI team can enforce it on a key.
When anonymization is active, does the model see my personal information?
No. In tokenization mode, each item is replaced with a consistent pseudonym, and the response comes back to you with the real values restored. In blocking mode, the request is refused at the first secret detected.
What happens if an item escapes detection?
Just before sending, a final check verifies that no original information survived. If one remains, the request is blocked. Likewise, if L2 detection is unavailable, the request is refused rather than sent with less protection.
Can I use only models hosted in Québec?
Yes. Sovereign models run on GPUs in Québec.
Can I keep my current subscription to an AI tool?
Yes, with the “anonymization only” mode: you anonymize with SecuAI, then send the text wherever you like.
Is a commitment required?
No. Prepaid credits and monthly subscription, with no term commitment.
Estimate SecuAI's monthly cost
The interactive simulator requires JavaScript. The prices above are authoritative.
From $29 per month, taxes extra — platform fee, modules extra.
| Module | Price |
|---|---|
| Anonymization (SecuAI) | Included |
Sovereign and market AI models: usage-based, drawn from prepaid AI credits. Per-model detail is in the simulator.
14-day free trial, once per product. No commitment. Price list 2026-09-24.2.
Estimate my cost — SecuAI
The interactive simulator requires JavaScript. The prices above apply.