Security and data

1. Isolation between companies#

• Every table that stores customer data has a company column and an active Row Level Security policy, in FORCE mode, which applies even to the table owner.
• The policy has a read rule and a write rule. Without the second, a company could write a row stamped as belonging to another.
• The application accesses the database with a low-privilege role, which has no permission to turn off the policy or to read another company's catalog.
• There is an automated suite that tries to cross the boundary between two companies and must FAIL. If someone weakens a policy, the test breaks before the deploy.

2. Who, inside your company, sees what#

• Permission by person and by role: who can see reports, who can configure the AI, who can touch billing. The default is the minimum, and you open up whatever you want.
• An agent can be limited to what is theirs, without seeing the whole inbox.
• The session expires, and the password is stored as a hash, never as text.

3. What gets logged#

The platform logs relevant actions with who performed them, and the "who" distinguishes a team member, an API key, the AI itself and the system. It is what lets you answer "who opened this conversation and when" without depending on memory.

An audit record is not deleted along with the business data when that would break the audit itself: the reference is undone and the event remains.

4. Where the files are stored#

Photos, audio and documents that pass through the conversation are stored in object storage on our own infrastructure, with each file's path stamped per company. Access to the bytes goes through the same session and company check as the screens, and not through a guessable public link.

5. Artificial intelligence#

When the AI is on, the conversation content goes to the model provider chosen in your company's configuration. We do not train models on your conversation. What the AI's internal tools (calendar, services) return carries no name or phone, and there is a test that breaks the build if someone puts identity there. Details and the list of sub-processors in the Privacy Policy.

You turn off the AI in the dashboard at any time. When it is off, no conversation content leaves for a model provider.

6. Export#

Your company's data can be exported while the account exists. If you need a specific format to migrate, write to [email protected] and we will agree on the format.

7. Delete#

You delete conversations and contacts in the dashboard itself, and request deletion of the whole account through the path in Data deletion. After cancellation, the data is kept for 30 days and is then deleted, except for what the law requires us to keep.

8. Found a flaw?#

Write to [email protected] with the subject SEGURANCA. We reply, fix it and credit whoever reported it, if the person wants. We do not take action against those who report in good faith and without causing harm.

9. What this page does NOT claim#

As a matter of principle, atendIA does not claim what it cannot prove. Today we do not declare ISO 27001 or SOC 2 certification, nor do we promise a contractual availability number. When any of these exists, it will appear here with the evidence. If your purchase depends on one of these items, talk to us before signing up.