Security written downfor your reviewers

Tenant isolation

Each customer gets its own logically separated workspace. Business units, regions or brands can have separate workspaces with their own data, playbooks and permissions.

Access and audit

Role-based access separates admins, managers and reps. Every action an agent takes is written to an audit log and to the CRM record it affects.

• Admins control who can edit playbooks and hand-off rules
• Managers approve new message templates before they are sent
• Reps see the full history of every lead handed to them

How agents are kept in bounds

Agents work inside limits your team sets before launch. Anything outside those limits goes to a person.

• Approved message library: agents send only messages your managers have approved
• Topic hand-off rules: topics you mark as sensitive, such as pricing or contract terms, always go to a person
• Human approval for risky actions: you choose which actions wait for a person to approve them
• Rate and send limits: caps on how much an agent can send, per lead and per day
• A rep's manual edit always wins: agents never overwrite it, and never negotiate, discount or commit on your behalf unless you allow it

How we test agents before and after launch

Testing is a practice that continues after launch, not a one-off check. It is not a certification.

• An evaluation set built from your real past leads, scored against the outcome your team wanted
• Changes to prompts or models ship only if they do not regress against that set
• Prompt-injection and data-leak testing, following the OWASP Top 10 for LLM applications as a practice
• Monitoring of quality, latency and cost per action once agents are live

Shared under NDA

Four topics are covered in our security pack, not on this page. Request it with the form below.

• Hosting and regions
• Model data use
• Subprocessors
• Certifications