A man monitors an AI chatbot workflow connecting messaging software to a secure cloud platform.
Nous Research has put its open-source Hermes Agent inside the Microsoft stack. On November 18, 2026, Nous Research, an M12 portfolio company and the team behind Hermes Agent, announced that the agent is available in native Microsoft services for the enterprise. The pieces are Hermes Desktop in the Microsoft Store, a Teams bot, and agent execution on Azure Container Apps Sandboxes. It landed during Microsoft Ignite, and Nous says it will present there in session BRK448.

Most of the claims come from Nous itself. This article separates what the vendor's own documentation supports from what is marketing, and ends with a pilot plan for Teams admins.

A man monitors an AI chatbot workflow connecting messaging software to a secure cloud platform. What was announced​

  • Hermes Desktop for Windows in the Microsoft Store. Nous's release doesn't include a managed-deployment guide. Check the Store listing and your own admin tooling before assuming it fits your software-distribution process.
  • A Teams bot. Employees hand off knowledge work such as financial analysis, document drafting or inbox triage, and review the result in the same conversation.
  • Azure Container Apps Sandboxes. Agent workloads run in isolated microVMs that start in under a second and release compute when idle.
  • Collective Wisdom. Employees propose and publish skills to their team, with review.
  • Hermes Enterprise. Nous says organizations with strict data-residency, security or procurement needs can run it in their own Azure tenant under a "similar product and pricing model."

Nous also claims Hermes is the top public open agent on OpenRouter and processes nearly 2 trillion tokens a day. Both figures come from Nous. I haven't independently verified either.

How the skills model works​

Nous describes Hermes as a self-improving agent. When an employee repeatedly delegates a task, Hermes turns the approach that worked into a reusable skill. Nous defines a skill as a written definition paired with code that runs the stable steps the same way each time. The claimed payoff is less inference and more reliability for established workflows.

Collective Wisdom extends this to teams. Nous says the next person to attempt a task starts from the best approach anyone on the team has found. It also says the resulting shared knowledge stays a private asset of the organization and isn't training data for an AI vendor.

That last statement is a marketing promise, not contract language. The release gives no detail on how the review step is enforced or who can see skills before they're published. Get both in writing.

Teams admin notes from Nous's help page​

Nous's Teams help page is the most useful document in this story because it is specific. It says the Teams app lets you chat with your own Hermes agent. Your agent runs on Hermes Cloud in the Nous Portal.

  • Permissions. Hermes is a bot. The page says it doesn't request Microsoft Graph or resource-specific consent permissions, and it doesn't sign users in to Microsoft.
  • What it sees. It receives messages sent to it in a 1:1 chat. In group chats and channels it receives only messages that @mention it.
  • Admin control. To allow it, open Teams apps > Manage apps in the Teams admin center, find Hermes Agent and allow it. App permission policies limit it to selected users, and app setup policies install it for them. To remove it from your organization, block the app on the same Manage apps page.
  • Linking. Each person who uses it connects their own Teams account to their own Nous account. The steps are:
    1. Add the app in Teams.
    2. Sign in to the Nous Portal and pick or create an agent.
    3. Turn on Team Gateway in the agent's settings.
    4. Click Link Teams.
    5. Send the generated link command to the bot in a 1:1 chat.
    The code expires after 10 minutes and isn't meant to be shared. To set up an agent for a teammate, create it for them in the portal and they get an invite email.
  • Disconnecting. To disconnect, send unlink to Hermes Agent in a 1:1 chat.
  • Limits. Files work only in 1:1 chats. Guests and people from other organizations aren't supported yet. Responses are AI-generated and may be incorrect.

A discrepancy on Microsoft Graph​

Some press coverage describes Teams features that go further than the help page does. One outlet says the bot answers direct messages, operates in channels and writes meeting summaries, reaching meeting content through Microsoft Graph. The help page, by contrast, says the documented app requests no Graph permissions.

The two may describe different configurations, or the coverage may be loose. I can't reconcile them from the available evidence. Before approving anything, ask Nous exactly which permissions each deployment option requests, and check the app manifest your tenant shows in the admin center.

Where your data goes​

Two deployments are being described, and they shouldn't be conflated.

The Teams app on Hermes Cloud. Messages, and files shared in a 1:1 chat, pass through the Nous connector service to your agent, together with your Teams account and organization IDs and your display name. The connector keeps a reference to each Teams conversation the app is in, plus the link between your Teams and Nous accounts until you disconnect. Conversations stay in your agent until you delete it.

Hermes Enterprise in your own Azure tenant. Customers can deploy it inside their own Azure tenant. The release doesn't say whether the Nous connector stays in the message path. It also doesn't say which Azure services are used, how updates are handled, or what contractual and compliance terms apply. Don't assume own-tenant hosting answers those questions. Ask.

What the Azure sandbox claim means​

Microsoft's own documentation backs up the platform claim in general terms. Microsoft Learn describes Container Apps Sandboxes as isolated environments for running code. Each runs in a lightweight microVM that starts in under a second. A stopped sandbox releases its compute resources.

Microsoft's reliability guide adds details that matter to anyone planning production use:

  • Memory mode preserves a stopped sandbox's full memory image and local disk. Disk mode preserves only the local disk, so processes restart on resume.
  • It's a single-region service. Sandbox groups aren't replicated across regions and there's no automatic failover.
  • It doesn't offer an availability SLA.
  • Microsoft recommends keeping durable data outside sandbox memory, for example in sandbox volumes backed by external storage.

This corroborates the platform's capabilities. It doesn't show how Nous configures its Hermes workloads. Isolation also doesn't make agent output correct. It limits where untrusted code can run, not what the agent decides to do.

Context: a partner announcement, not a Microsoft product launch​

Nous identifies itself as an M12 portfolio company, and M12 is Microsoft's venture fund. The release quotes M12's Michelle Gonzalez and Nous CEO Dillon Rolnick. This is a partnership announcement from a portfolio company at Microsoft's conference. It isn't evidence that Microsoft has validated the product's performance or built it as a Microsoft service.

Parts of this aren't brand new. One outlet reports that Nous had already made Hermes Agent available to Windows users through the Microsoft Store before this announcement. The same outlet notes the agent runtime has been under the MIT license since February 2026. The new elements are the enterprise framing, Collective Wisdom and the own-tenant option. Hermes also now competes with Microsoft's own agent offerings inside Teams and Azure, so expect overlap in procurement conversations.

What isn't published​

  • Any price. "Similar product and pricing model" is the only statement.
  • Tenant deployment details.
  • Certifications, audits or data-processing terms.
  • A named enterprise customer or measured result.
  • A deployment guide for managing the Store app.

A two-week pilot for admins​

This plan is my suggestion, not a Microsoft or Nous procedure. It uses only features Nous describes.

  1. Start with a restricted group. Allow the app in Manage apps, then use app permission and setup policies to give it to a small pilot group.
  2. Pick three checkable tasks. Weekly reports with fixed fields work better than open-ended drafting.
  3. Set a baseline. Record time saved, error rate against a human version, and how often a person had to redo the work.
  4. Test shared skills. Have one person delegate to Hermes for several rounds. Then have a second person use the resulting skill through Collective Wisdom and compare results.
  5. Read every skill before publishing. Review both the written definition and the code.
  6. Check the data path. Send a harmless marker string and ask Nous where it is stored and for how long. Compare the answer with the help page.
  7. Decide on the tenant option. If your compliance team requires residency, get the own-tenant terms and pricing in writing.
  8. Keep a human in the loop. Review anything that leaves the team, such as a customer email.

Bottom line​

This is a real, documented integration. The help page gives admins concrete permission, scope and control details. The Azure sandbox claims are consistent with Microsoft's documentation. The scale claims, the data-use promise and the own-tenant architecture rest on Nous's word alone. Run a small pilot, and get the data path and tenant terms in writing before widening access.

 

References

  1. Hermes Enterprise in Microsoft Teams: What Nous Announced - explainx.ai explainx.ai 2026-11-19T00:00:00+00:00
  2. Nous Portal portal.nousresearch.com
  3. Nous Research partners with Microsoft to bring Hermes Agent to the enterprise nousresearch.com