Work / Risk and compliance · 2026
An AI assistant that works inside the user’s permissions
An enterprise AI assistant inside a compliance platform, running on each customer’s own Claude or OpenAI key rather than a shared one.
It answers questions about the screen someone is on and carries out work through the platform’s own APIs, under that person’s permissions, with every call written to an audit log.
In production on each customer’s own model keys, every tool call logged
What’s in it
- What it does
- Permissions are the boundary
- The customer brings the key
- Everything is logged
- What’s actually hard