Use AI without exposing your data.
Consumer assistants expose what you paste into them. iYan keeps the whole chain under your control: the request is anonymised, processed in a perimeter you chose, answered from your own sources, and recorded.
Canada
data residency
AES-256
encryption at rest
100 %
of requests traceable
1
platform instead of five tools

Three risks that public AI tools create.
They are not theoretical. They are the questions your customers, your auditor and your privacy officer will ask.
Customer data sent abroad
Even with a "no training" switch, a public assistant processes your prompts outside Canada, under laws that can compel access.
Compliance you cannot prove
Law 25 and GDPR expect a register, a lawful basis and the ability to erase. A public tool gives you none of that for what your teams paste.
Shadow AI
When the official tool is missing, people use their own. The data is then spread across accounts nobody manages.
Six protections, on by default.
No configuration needed to be protected; every control can be tightened further by an administrator.
Anonymisation before the model
Personal information and your own referentials become anonymous tokens before any model reads the request. The answer comes back with the real values.
Encryption everywhere
TLS in transit, AES-256-GCM at rest for the vault, a master key in a secrets manager, encrypted backups. Organizations can hold their own keys.
Fine-grained permissions
Rights per workspace, folder, file, board and connector. Invitations, expirable public links, a per-user identity for every connector.
Answers from your sources
The knowledge base indexes your documents locally, cites them in each answer, and keeps versions. Web search and external connectors can be turned off.
Choice of where the model runs
Local or private models for sensitive work, cloud models only when you approve them. The allowlist is set per organization.
A complete audit trail
Who asked what, which tool ran, which file was read: every event is hashed into a chain and exportable. Alerts flag sensitive actions.
The model never sees the real values.
What anonymisation covers.
- 1
Identities
First and last names, usernames, employee numbers.
- 2
Contact details
Emails, phone numbers, postal addresses.
- 3
Identifiers
Account numbers, file numbers, health or social insurance numbers, card numbers.
- 4
Your referentials
Customer, supplier, employee and product lists you import: every occurrence is masked, including inside documents.
1. Original text
Marie Tremblay (marie.tremblay@example.ca) asks for a refund of invoice F-2031-77.2. Sent to the AI engine
[PERSON_1] ([EMAIL_1]) asks for a refund of invoice [REF_1].3. Restored answer
Refund of invoice F-2031-77 approved for Marie Tremblay. Confirmation sent to marie.tremblay@example.ca.
Real values stay in your workspace; the model only receives placeholders, and the answer is rebuilt on the way back.

The same work, without the leak.
An analyst summarises a contract, a clinic drafts a letter to a patient, an HR team compares applications. In iYan the document stays in the encrypted vault, the sensitive values are masked, the model answers, and the real names reappear only on the user's screen.
Administrators decide which models are allowed, whether web search is on, whether masking can be turned off, and for how long the audit journal is kept. In the HIPAA mode these choices are locked to the strict value.
When an auditor asks, the journal answers: the request, the files read, the tools called, the person, the time, and the hash that proves nothing was altered.
From the question to the answer.
Five steps, each one logged.
- 1
Authenticated request
SSO or password with 2-step verification; role and permissions checked.
- 2
Local processing
Personal information masked, referentials applied, size and content limits enforced.
- 3
Approved model
The gateway routes to a model allowed for your organization, with your quota and your keys.
- 4
Answer from your sources
The knowledge base grounds the answer and cites the documents used.
- 5
Signed journal
Request, response, tools and files recorded in the tamper-evident chain.
Frequently asked questions
How is data residency guaranteed?+
By the architecture, not only by contract: the platform and its storage run on servers in Canada, or on yours. Model calls go only to providers you allowed; local models keep everything inside the perimeter.
Does iYan comply with Law 25, GDPR and HIPAA?+
Law 25 and GDPR obligations are implemented: person in charge, register, rights of access and erasure, incident notification. HIPAA is an activatable mode that locks the strict settings and adds a BAA registry; it requires an organization plan.
What is retained, and for how long?+
Your content stays as long as your account exists and is erased on request. The audit journal is kept for the period your sector requires, six years in the HIPAA mode. Backups are encrypted and rotated.
Can my team still use web search or external connectors?+
Yes when you allow it. An administrator can disable web search, restrict connectors, or let the HIPAA mode disable both.
How are access rights managed?+
SSO (OpenID Connect, Google, Microsoft, Apple), 2-step verification that an administrator can make mandatory, idle sessions that expire, roles per workspace, per-file sharing, and a journal of every access.
What if a connector or an agent misbehaves?+
Connectors run in a sandbox under a per-user identity, with only their own folder visible. Agents work with a step budget, ask before sensitive actions, and are stopped when they loop.
Deploy AI without a data risk.
We configure the perimeter, the allowed models, the masking rules and the audit retention with your team.