When you hand company data to an AI, "trust us" isn't enough. No marketing on this page: where your data lives, who can reach it, and what the assistant cannot do.
Every tool call passes a code-level policy engine BEFORE it runs. This is not a prompt rule — prompt injection cannot override it. Risky actions fall to your approval.
Your knowledge base and your chats are never used to train any AI model. Traffic to AI providers goes through our own gateway to enterprise accounts that do not train on it.
Every tenant's data is tagged with its own identity; every API request and vector search passes a tenant filter. Your knowledge base and your conversations are never shared with another customer.
Every tool call passes a code-level policy engine before it runs: approval thresholds, domain allow-lists, and a "never-silent" action class. This is not a prompt rule — prompt injection cannot override it.
What we hold is listed plainly in the panel: conversation records, knowledge base content and connected store details. Delete what you want with one click; on account closure everything is purged for good.
Servers hosted in data centres in France (EU). All traffic encrypted end-to-end via Cloudflare.
Your store and email connection credentials are stored encrypted. Secret values never reach the LLM and are masked in logs.
Two-factor authentication (2FA), role-based access control (RBAC), session security and failed-login protection are standard.
Untrusted content pulled from the web passes a dual-LLM quarantine: the quarantine model extracts only goal-relevant facts and strips instructions buried in the page, so external content can't steer the assistant.
The database runs in a high-availability cluster (automatic failover within seconds); the app is multi-node. Regular backups with restore drills; 24/7 health checks wired to alerts.
Crawl decisions and store actions are logged: what ran, why, with whose approval — fully traceable.
Every file a customer attaches in the chat, on WhatsApp, on Instagram or by email passes a virus scan and can't be downloaded until it comes back clean. Logos and knowledge-base files your team uploads are scanned after upload too: a harmful file is removed and written to the Audit Log. The uploaded knowledge-base file itself is deleted once its text has been extracted.
We use a limited number of sub-processors to deliver the service. The main ones:
| Sub-processor | Purpose | Data |
|---|---|---|
| Contabo (France, EU) | Server hosting | All application data (on encrypted disk) |
| Cloudflare | CDN, TLS, DDoS protection; file storage (R2) | Traffic (encrypted in transit); uploaded files (conversation files in private storage) |
| AI providers (Anthropic, OpenAI, Google, Together — via our gateway) | Model inference | Query context (never used for training) |
| Resend | Transactional email | Email address, notification content |
Think you found a security issue? Write to us directly — we respond within 48 hours.
[email protected]If your security team has a checklist, we'll gladly answer it — including DPA and data-handling details.
Schedule a security call