Short answer
Hermes Agent is Nous Research's free, open-source, self-hosted AI agent that remembers past sessions and builds its own skills. As of October 2026 it is excellent for technical people who want control and no license fees. For a typical owner it means installing, securing and maintaining software, and phone calls and team features are not confirmed.
What Hermes Agent is
Hermes Agent is an open-source (MIT) agent from Nous Research that you run yourself. It is described as self-improving: it keeps memory across sessions, creates reusable skills after complex tasks, and has a built-in cron scheduler, which means it can run jobs on a timetable without you present. Most sources date its launch to February 25, 2026, so say early 2026.
It runs on Linux, macOS, WSL2, native Windows, Android through Termux and Docker, and can sit on a small VPS or serverless backends such as Modal or Daytona. A messaging gateway reaches Telegram, Discord, Slack, WhatsApp, Signal, email and a command line.
What it does well for a business
- No license fee. The software is free. You pay for hosting and for the models you connect.
- Model freedom. It works with Nous Portal, OpenRouter, OpenAI, Anthropic and custom endpoints, so you are not tied to one provider.
- Real memory and skills. It learns from complex tasks and reuses what worked.
- Unattended jobs. The cron scheduler suits recurring work such as daily reports.
- Your data stays where you put it. It is self-hosted, so you decide where it lives, although data handling still depends on the model provider you connect.
For a developer or a technical founder, that is a strong package, and the price is hard to argue with.
What to weigh before using it
| Question | What we know (October 2026) | What it means |
|---|---|---|
| Who sets it up? | You, or someone you hire | Time or contractor cost, plus ongoing upkeep |
| Who maintains it? | The operator; self-hosted and operator-managed | Updates, security and uptime are your job |
| Business phone line | Phone calling not confirmed; voice-memo transcription and text-to-speech are listed | Look elsewhere if calls matter |
| Team features | Not confirmed | Shared memory, roles and approvals need design |
| Hosted service | A hosted Hermes agent service from Nous is not confirmed | Nous Portal is a separate paid subscription for models and tools |
| Compliance | No compliance claims in the sources we checked | Do not assume HIPAA or SOC 2 support |
None of this is a flaw. It is what self-hosted means. The hidden-cost list for this route is in hidden costs of DIY AI agents.
Who should use Hermes Agent
It is a good fit for an owner who is also an engineer, for a company with technical staff and unusual data needs, or for someone who wants to experiment and learn how agents work. If you enjoy a weekend of setup, you will probably like it.
It is a weaker fit for an owner whose time is the scarce resource. A real business workflow needs connections to your calendar, email and CRM, safe handling of logins, approvals before anything is sent, and someone to fix it when a service changes its login page. Those are all things you would build.
What a first month with Hermes looks like
- Pick a host: a small VPS, a spare Mac or Linux box, or a Docker setup.
- Install it, connect a model provider and decide which messaging apps it may use.
- Give it one narrow job, such as a daily summary of new leads, and let the cron scheduler run it.
- Review what it did, correct it, and let it build a reusable skill.
- Only then add anything that touches customers or money, with approvals on.
Treat inbound messages as untrusted input and limit what the agent can reach. That is good practice for any self-hosted agent, and it is the part non-engineers most often skip.
The managed alternative
Oraculo is the managed version of the same promise: an always-on agent that remembers and works across your tools. An engineer connects everything with you in four 1-hour Zoom sessions, and from then on you delegate by message on WhatsApp, iMessage, Slack, Telegram, email or the app.
It answers and places calls on your business line, follows up with leads, builds pages, sends invoices and keeps staff on their goals. Memory is shared across every channel, and important actions can require your approval first. Data is encrypted in transit and at rest, with role-based, logged access, and you can export everything and leave.
Starter is $12 per hour of work with a 6-hour weekly minimum, about $312 per month, plus a $499 one-time setup and a 30-day money-back guarantee. The two-way view is Oraculo vs Hermes Agent, and you can book a 20-minute demo.
Our verdict
Hermes Agent is good for what it is built to be: a flexible, free, self-hosted agent for technical people. It is not designed to be a ready-made employee for a business owner. Try it if you want to learn. Pay for a managed service if you want outcomes this month.
Key takeaway
Hermes Agent is a strong free tool for technical users. If you want an agent that arrives set up, answers your phone and reports to you, a managed AI employee is the more practical route.
Frequently asked questions
Is Hermes Agent free to use?
The software is free under the MIT license. You pay for your own hosting and model usage, and Nous Portal is a separate paid subscription for models and tools. A hosted Hermes service is not confirmed.
Can Hermes Agent answer business phone calls?
Phone calling is not confirmed. The listed voice features are voice-memo transcription and text-to-speech. If you need your phone line answered, choose a product that lists it, such as Oraculo.
Do I need to be technical to run Hermes Agent?
In practice, yes. It is self-hosted, so you install, secure and maintain it. A non-technical owner can hire help, but the ongoing upkeep stays with whoever operates it.
Does Hermes Agent work on WhatsApp?
Its messaging gateway lists Telegram, Discord, Slack, WhatsApp, Signal, email and CLI. Setup and access control are the operator's job.
Is Hermes Agent safe for client data?
It is self-hosted, so you control where data lives, but security depends on how you configure it and which model provider you connect. No compliance claims were found in the sources we checked.