Built around controlled access.

Workspace boundaries

The backend scopes agent configuration, provider settings and knowledge articles to the customer workspace. Access checks protect those business records, with tests for cross-workspace access and owner restrictions.

Provider credentials

BYOK provider settings use encrypted storage for saved telephone and AI credentials. Secret values are excluded from displayed forms and serialised records. Saving credentials does not mean a provider connection has been verified or live calling is enabled.

Approved business knowledge

Knowledge articles support draft and approved states so teams can prepare the information intended for their assistant. Live-agent retrieval and external data ingestion are still in development. Review articles before approving them and keep credentials out of content.

Permissions and caller verification

The live integration roadmap requires server-side enforcement of permitted tasks and caller verification before sensitive information is disclosed or changed. A caller knowing a name, email or telephone number is not by itself proof of authorisation.

These live-call controls must be implemented and tested before integrations handle customer requests. The marketing examples illustrate the intended experience.

Human fallback and data retention

Human transfer, provider failure handling, recording notices, transcript controls and data retention are part of the live voice roadmap. Their production behaviour is not yet verified. Businesses will need a tested escalation route before an assistant takes live calls.

Deployment and assurance

Production HTTPS, operational monitoring, backups, restoration checks and provider validation remain deployment requirements. We do not currently claim independent security certification or a completed production security review.

Report a security concern

Use our contact page or telephone number to report a concern and request a suitable channel for further details. Include the affected feature and a general description, but do not send passwords, API keys or customer records in an initial enquiry.

Separate instructions from permissions

An agent prompt describes intended behaviour, but it cannot serve as the only protection for customer records. The live integration service needs to verify access on every sensitive tool request, including the workspace, caller state and permitted action. External articles, tickets and store content must be treated as information rather than new permission to change system behaviour.

Prepare a sensible review routine

Keep an owner for each knowledge article and agent role. Review public policies when they change, remove unnecessary information and confirm that escalation instructions are still useful. In BYOK workspaces, use authorised provider accounts and arrange credential replacement when a key is no longer appropriate. Saving a new key still requires connection validation.

A production release needs operational evidence

Access tests are one part of assurance. Live-call behaviour, provider outages, usage accounting, backups, restoration and monitoring need to be checked in the deployed environment. The security overview should be updated when those checks establish real production behaviour, with the remaining limits clearly described.