Grounded answers
The knowledge role answers only from documents found for the question. No answer means an honest handover to your team.
Security
Your customer conversations are the asset. We treat them that way.
This page describes controls implemented today. We state current limitations instead of leaving you to assume.
Workspace isolation
Every customer record, conversation, message, document, business record, appointment and team member belongs to one workspace. Isolation is enforced by database row-level security; browser sessions never receive privileged credentials. Automated tests cover cross-workspace isolation on every change.
Credentials and secrets
Connection credentials are held in encrypted server-side secret storage. They are never exposed to the browser or committed to source control. Payment and email keys are restricted to the minimum required scope.
AI agent behaviour controls
Untrusted input
Customer messages and documents are treated as data. Instructions inside them cannot change an AI agent's role, reveal configuration or trigger actions.
Mandatory escalation
Requests for a person, complaints, anger, legal or financial questions, and repeated low confidence pause the AI and notify your team.
Change history and auditability
Changes to workspace settings, team access, AI agent instructions and operating rules are recorded with previous and new values. Instructions are versioned, and every run records response time, cost and outcome. Data changes are reviewed and tested before release.
Application hardening
The application uses HTTPS, strict transport security, a content security policy, restricted cross-origin access, validated redirects, and upload limits. Every privileged server action checks workspace membership and role, and the build enforces that requirement automatically.
What we confirm during onboarding
Data region
We confirm the hosting and processing region that applies to the workspace in writing instead of implying an unverified residency claim.
Subprocessors
We identify the providers involved in the workflow, including infrastructure, AI model services, messaging channels, calendar, and CRM systems.
Pilot scope
We avoid collecting more sensitive data than the pilot needs. Data sample, channel, integrations, and success target are scoped before activation.
Current limitations, stated plainly
- Detoximex does not currently hold SOC 2, ISO 27001 or equivalent third-party certification. We do not claim one.
- Message content is processed by our AI model service to generate replies. Do not load a dataset into the knowledge base if that processing is unacceptable.
- Data is hosted with infrastructure services; the region and retention terms for your workspace are confirmed in writing during onboarding.
- Staff replies sent from the workspace are delivered by the recovery cycle and can take up to 15 minutes to leave the platform; they are refused once 24 hours have passed since the customer last wrote.