Hermes Agent
Learns as it goes
Hermes Agent is a self-improving open-source agent from Nous Research. It turns experience into reusable skills and remembers across sessions.
- Cost
- Free softwareplus a server and model costs
- Setup, roughly
- A day or more
- Runs on
- Your own machine or a VPS
- Works on
- Mac, Windows or Linux
Checked 7 Nov 2026.
You Check these sources and tell me only what has changed since last week. If nothing has, say so in one line.
Two things changed since last week.
- Northwind raised its prices by 4%On their rate card, updated FridayNew
- A rival opened a second storeLocal news, WednesdayNew
- Everything else stayed the sameSix other sources checked
You Draft a short summary of this week's open issues for our team channel. Show it to me before you post.
Five issues open. Two need a decision.
- Two orders are waiting on stockDecide: reorder or substituteDecide
- The new starter still needs a loginAsked IT on TuesdayDecide
Our take
Hermes Agent gives you flexibility and control, and it improves its own skills as it works. Like any self-hosted agent, it needs someone who can secure it.
Our stake: Navigate helps clients set up agents like this one. Nobody paid for this guide.
What is Hermes Agent best at?
- Scheduled researchRuns on a timer and reports back.
- MonitoringKeeps an eye on what you care about.
- Skills that improveCreates skills from experience.
- Your chat appsWorks across chat apps.
How much does Hermes Agent cost?
Model costs depend entirely on your use. Prices are in US dollars. Checked 7 Nov 2026.
How do I set up Hermes Agent safely?
8 steps. Allow a day or more, then a week of checking. We put the approvals on before anything else.
- Use a dedicated machine
- Set a spend limit first
- Use the Docker backend
- Keep the dashboard local
- Approve skill changes
- Narrow each scheduled job
- Give it your house rules
- Know how to stop it
See every step
1. Use a dedicated machine. A small rented server (VPS) is enough to start. Keep it separate from your main laptop.
2. Set a spend limit first. Before it runs anything, set a monthly spend limit with your model provider, or load only a small prepaid balance.
3. Use the Docker backend. It limits what the agent can touch.
4. Keep the dashboard local. Reach it over an SSH tunnel, not the open internet.
5. Approve skill changes. Turn on write approval so its self-edits stay reviewable.
6. Narrow each scheduled job. Give each job only the tools it needs.
7. Give it your house rules. Paste these in before it does any real work.
8. Know how to stop it. Stop the agent on its machine to halt everything, and revoke its model key if you need to cut it off fast.
Is Hermes Agent safe?
Only as safe as the machine you run it on. This is what we would watch.
- Security is on youUse the Docker backend, and keep the dashboard local.
- It edits its own skillsTurn on approval for skill changes so they stay reviewable.
- Fast releasesCheck what changed before you upgrade.
Who should skip Hermes Agent?
- A pipeline must run the same way every time
- You don't want to run and secure it yourself
- You work somewhere an agent changing its own skills is a liability
How has Hermes Agent done in tests?
Research benchmarks such as WildClawBench and Harness-Bench have tested Hermes Agent itself, alongside OpenClaw, Claude Code and Codex. The best result depends on the model, so none of them wins outright.
- Personal agentsPersonalAgentBench, by micro1. Not tested
- Coding agentsAA Coding Agent Index, by Artificial Analysis. Not tested
- Self-hosted agentsWildClawBench, by InternLM. Tests this product
- Office tasksGDPval-AA, by Artificial Analysis. Not tested
- Running a businessVending-Bench, by Andon Labs. Not tested
- Freelance workRemote Labor Index, by Center for AI Safety and Scale Labs. Not tested
Alternatives to Hermes Agent
The details
Good to know
Who runs it. You do. Its docs cover Linux, macOS, WSL2 and native Windows.
Model providers. Nous Portal, OpenRouter, OpenAI or any endpoint.
Licence. MIT.
Sources for this page
Hermes Agent docs (Nous Research), read 7 Nov 2026
WildClawBench, read 7 Nov 2026
Harness-Bench (arXiv), read 7 Nov 2026
Know better?
Tell us what we've missed at info@navigateagents.com. We'll fix it and credit you if you'd like.