The browser is the gate
Syncanix’s security model in one idea: the AI never holds your users' credentials.
The most important thing to understand about Syncanix is what the AI never sees. The assistant works from a structured description of your capabilities and the user’s request — not from your source code, your secrets, or your users' credentials. Identity and access stay where they belong: with your users and your identity provider.
Structured data, not source or secrets
The language model receives the capability catalog (names, descriptions, input shapes) and the conversation. It does not receive your source code, your environment variables, your database, or raw request and response bodies. During discovery, only structured metadata leaves your machine — and only after you consent.
Identity stays at the edge
When an action runs, the end user’s identity comes from your identity provider at that moment, in the user’s own browser or client. The assistant never stores or replays a credential. It can only ask to perform an action; your systems decide, using the user’s real identity and permissions, whether it happens.
Why this is safe
- An action can never exceed the user’s own permissions — even a fully manipulated prompt can only attempt what that user could already do.
- Actions that change or delete data are gated by confirmation, and destructive actions are blocked unless you allow them.
- Every action is recorded in an audit trail, attributable to the user who ran it.
Privacy during discovery
Discovery is static by default: it reads your code without running it. Optional AI enrichment only runs after explicit zero-data-retention consent, and a .syncanixignore file lets you exclude anything you don’t want scanned. File paths, line numbers, secrets, and your data never leave your machine.