Provider keys that never reach the browser, prompts cleaned before they are sent, one control point for everything that leaves for a model provider, and tenants sealed off from each other.
Six protections that work underneath every AI screen — no per-screen set-up to forget.
Every call to a cloud model is made by the TotalApp server, never by the browser, so there is no provider key in the browser to steal.
HTTPS only with a one-year HSTS policy, signed tokens that expire after 15 minutes, an HttpOnly refresh cookie and per-route rate limiting.
When switched on, prompts are cleaned of hidden characters, script blocks and known injection phrases — stripped and continued, or blocked outright.
All outbound AI traffic passes one control point on the server, so masking and topic rules cannot be forgotten on a single screen.
Uploads are checked against server-side type allowlists and their text is extracted without ever being executed.
Data is stored per tenant and the tenant is taken only from the signed token, never from a request field.
Choosing a local path is the strongest data-handling option: the text never leaves your environment.
| Path | Where your text goes |
|---|---|
| Local Ollama (chat or embedding) | Browser to your own machine. Nothing reaches TotalApp or a provider. |
| In-browser models | Processed inside the browser tab. Nothing leaves the device. |
| Cloud LLM, hosted embedding, rerank | Browser to the TotalApp server over HTTPS, then to the provider you chose. This is the path the egress controls govern. |
The control sits on the server’s transport layer, so it applies to every AI feature without each screen opting in.
The request is matched against the AI endpoints the server uses — chat, embeddings and rerank. Anything else is left alone.
The tenant and user of the signed-in request are identified and that tenant’s policy is applied. Every control is off until an administrator turns it on.
Prompt text is cleaned, personal data is masked and blocked topics are checked. A blocked request is never forwarded.
The answer is masked and checked the same way before it is returned. Streamed responses pass through unmodified.
If the audit trail is on, a metadata record is written in the background.
| Risk | Protection |
|---|---|
| Credentials exposed to the browser | Provider keys exist only on the server |
| Traffic intercepted or session hijacked | HTTPS everywhere, short-lived signed tokens |
| Instructions smuggled into a prompt | Input sanitization and injection detection |
| Personal data sent to a model provider | A single egress control point with masking |
| Unsafe uploads | Server-side type allowlists, text extraction without execution |
| One tenant reaching another | Per-tenant storage with path containment, tenant taken from the token only |
AI Governance is the set of rules your administrators choose — permissions, blocked topics, masking policy and audit export. AI Security is the technical protection underneath that enforces them and protects the platform.
The tenant policy controls — prompt sanitization, masking, blocked topics and the audit trail — start switched off and are enabled by a tenant administrator. Platform protections such as server-side provider keys, HTTPS, signed tokens and tenant isolation are part of how TotalApp works.
No. Detection is pattern-based: it removes well-known attack shapes and hidden-text tricks but cannot recognise every novel phrasing. Treat model output as untrusted text — TotalApp displays answers, it does not execute them.
Yes. With local Ollama or in-browser models the text never reaches TotalApp or a provider.
The egress control sits on the server’s outbound AI traffic, so it covers every AI feature that calls a model provider without each screen opting in.