From the source
Nous Research is adding a capability to its open-source Hermes Agent that lets it work as an AI coworker with its own work account, run on
Microsoft Foundry.
- The company says the change answers a question security teams ask about enterprise agents: whose credentials is that agent using?
- Today most enterprise agents borrow a person's identity through an OAuth consent screen or an MCP server, acting under that employee's name while the browser tab stays open.
- Nous Research says that approach forces IT to build a second control system, confuses employees who cannot tell what they allowed, logs the agent's actions under the user's name, and leaves software running unattended with full access to email, files and chats.
- Under the new model each agent gets its own email, cloud drive, calendar, chat profile and manager, so IT governs it with the tools it already uses for staff and employees share a folder with it as they would with a colleague.
- Microsoft Foundry runs the agents inside the company's Azure subscription, giving each session an isolated virtual machine, shutting the agent down when it is idle, and supplying built-in tools such as web search, a hosted browser and a Code Interpreter.


