An endpoint agent reduces AI data leakage by inspecting traffic to AI services locally, at the device level, rather than relying on a browser extension a user can disable or a proxy that has to be manually configured on every machine.
How it's different from a browser extension
SafePrompt used to ship as a browser extension. It's now a native agent that locally intercepts connections to known AI-service domains (ChatGPT, Claude, Gemini, Copilot) using TLS interception on the device, and inspects the content before it's forwarded. That means coverage isn't tied to Chrome, Edge, or any single browser — it works regardless of which browser (or app) the request came from.
What it catches
- Credentials — API keys, passwords, database connection strings, SSH keys
- PII — names, emails, phone numbers, SSNs, credit card numbers
- File uploads — PDF, DOCX, XLSX attachments to AI chat tools (Professional and above)
API-based detection
Professional and above also expose a local detection interface — other applications on the same device can submit content to the agent's detection engine directly, rather than only relying on network-layer interception. This is for internal apps or integrations, not a hosted management API.
What's not shipped yet
Coverage for Microsoft Office add-ins, Outlook plug-ins, and desktop AI copilots/assistants would need UI Automation or clipboard-level interception, which isn't built yet. We're evaluating it based on customer and market research, but it isn't committed or scheduled — we won't rush it out at the expense of a proper security review. We'll update this page if that changes.