🤖 Agent PoC

What this PoC demonstrates

This is a standalone demo of a role-aware assistant backed by real enforcement in code — not just a model asked nicely to behave. Three things to try, each with its own angle:

🔒 Security — permission segmentation

Switch to Bob or Carol (regular users) and try an admin-only action, like deleting a record or viewing the audit log. It gets refused — the check happens in the code (data.js/the active MCP server), never just in the model's judgment.

🕵️ Privacy — multi-user isolation

Each user only ever sees their own records — conversation history and data are never mixed between users. Try asking Bob for a record that belongs to Carol: he's refused without the reply even confirming the record exists or who owns it.

🌐 Gateways — same backend, three doors

This same chat, the REST API (POST /api/chat), and Slack (if configured) all talk to the exact same agent and enforcement — nothing is duplicated per channel.

Not sure what to type?

Use the suggestion chips above the message box — they change based on which user is selected. Or just ask the agent directly: "What can you do?" or "How do I test this?" — it can explain itself and suggest things to try.