Security and data handling

• Your accounts. The code repository, the hosting, the database, and the logs are in accounts opened in your name. I am added as a user. I do not keep a copy in mine.
• Your model keys. Calls to Claude or OpenAI go through an API account in your name. That provider's retention and training terms apply to you directly, and you can read them yourself: Anthropic, OpenAI.
• Only what the task needs. Each step sends the model the fields it requires, not the whole record.
• The action log. Every read, decision, change, and approval is written to your database. It is yours to export or delete.

• Production access is named. It is me, through your accounts, listed in the engagement agreement. You can remove it at any time.
• No shared credentials. Each system gets its own key, scoped to the minimum it needs, and the keys live in your hosting environment rather than in the code.
• Approvals are yours. Anything with a price, a promise, or a write to inventory or a ledger waits for a person you name. I do not approve on your behalf.
• There is an off switch. One setting pauses the automation without touching your other systems.

Stated plainly so you can decide whether this fits before we talk.

• No SOC 2 report. I am one person, not a certified organization.
• No contracted backup developer. The runbook, the recorded walkthrough, and client-owned accounts exist so another developer can take over, but a named second person is not on retainer.
• If your work is covered by client-confidentiality duties, ask your own counsel before sending me client documents, and expect to review the data flow for your specific system during the architecture sprint.

If you find a security issue on this site or in something I built, email contact@eyalgantz.com. I read it the same day.