Hermes takes on a case. You stay in control.
We install Hermes, the open source AI agent by Nous Research, on a server you own. We give it a precise mission, written down, and wire it into your tools the way we would integrate a new service.
- Ticket receivedsupport@ · "Duplicate invoice for March"
- Context foundCustomer account, 2 identical invoices on 03/03
- Cause identifiedPayment triggered twice, credit note needed
- Credit note and reply draftedRefund: human approval required
What is Hermes?
Hermes Agent is an open source AI agent developed by Nous Research. It runs continuously on a server, keeps a memory of its work, uses tools (messaging, files, browser, APIs) and runs scheduled tasks.
It receives an event, such as an email, a ticket or a form, and moves the case forward within the limits we set, without waiting to be asked.
Missions we give it
First-line support
It reads the request, finds the customer context, answers simple cases and prepares a summary for the team on the rest.
Sales monitoring
It watches useful signals (hiring, funding, tenders), qualifies accounts and drafts approaches for review.
Admin follow-up
Missing documents, pending files, deadlines: it chases, updates the internal tool and flags what is stuck.
The mission sheet, before any configuration
An autonomous agent without a written frame always ends up surprising you. This is the sheet we fill in with you for a support agent.
- Mission
- Handle first-line billing tickets arriving at support@.
- Allowed sources
- Support inbox, CRM (read), billing tool (read), knowledge base.
- Acts alone
- Answer procedure questions, resend an invoice, tag and file the ticket.
- Needs approval
- Credit note, refund, commercial gesture, reply to a customer in dispute.
- Forbidden
- Change a price, delete data, email a customer outside a ticket.
- Reporting
- Daily summary in Slack, full action log available.
When Hermes is not the right choice
If the rule is simple and stable, an n8n workflow does better for less. If you mainly want an assistant to ask questions in your messaging app, look at OpenClaw.
Questions about Hermes
Can Hermes act on its own?
On simple cases allowed by the mission sheet, yes. On a sensitive message, a refund or a delicate reply, it prepares and a human approves.
Does Hermes learn over time?
It keeps a memory and can save reusable procedures from what it has already done. We review what it saves so its behaviour stays the one you approved.
Which AI model does it use?
Hermes works with several model providers. We choose based on the task, cost and your data constraints, and can switch without rebuilding everything.
What if a tool changes its API?
The agent fails cleanly and reports it. As part of ongoing support, we fix the integration and check the mission before restarting it.
A mission to hand to an agent?
Describe the work and what must never go out without you. We'll tell you whether Hermes holds up.