I’d like to share a new NS8 module I’m working on: Hermes Agent.
What is Hermes Agent?
Hermes Agent is an open-source autonomous agent from Nous Research. It is designed to run on your server, keep context over time, and become more capable as it learns, rather than being just a local chatbot or IDE copilot.
Why I made this NS8 module
I built this module because I wanted the easiest way to deploy one or multiple Hermes agents on NethServer 8. Every istance of the module manages one or more Hermes Agent runtimes.
The goal is to make setup and day-to-day management much simpler for NS8 users.
How to try it
Install it:
add-module ghcr.io/stell0/hermes-agent:0.1.0 1
Than from NS8 UI configure
a virtualhost for the agent dashboard like hermes.example.com
select a user domain from the dropdown, which binds the module to that domain and populates the allowed_user selectors for each agent
one or more agents with unique allowed_user values from the selected user domain
Configuration will create the agents and publish the dashboard at https://hermes.example.com/ with per-agent authentication and routing.
From dashboard, you can setup a Telegram and everything else.
multiple users for one bot, and multiple bots for one user
easier integration with other NethServer 8 modules
multiple agent profiles for each agent
Known issues
the Dashboard Web UI is build every time the container starts, so it takes a bit of time to be available after the agent service is started.
the module does not support multiple agents with the same allowed_user value.
after changing the configuration from dashboard, the agent service needs to be restarted to apply the new configuration. At the moment it can be done with the /restart command, but the first time you coinfigure a messaging platform you need to restart the service from terminal with systemctl --user restart hermes@<id>.service or saving changes from NS8 ui
At the moment, saving changes from NS8 UI restart all the agents, but in the future we will implement a smarter logic to restart only the agent that needs it.
Feedback and contribution
Feedback, testing, ideas, and contributions are very welcome.
If you try it, I’d be interested in feedback on the deployment flow, the UI, and how this should integrate with the rest of the NethServer 8 ecosystem.
Thank you for providing this.
Quick note: The installation went smoothly and I’ve managed to set up the first agent (Developer). I can access the dashboard.
I can’t say much about using it yet, as I still feel a bit unsure. I have no experience with agent-based AI and am simply curious.
I’m now looking forward to seeing what experiences are shared here in the community. I hope my appetite will grow as I go along.
I don’t know if my hope is justified. But I wish there were a way to get a consolidated analysis of the various DMARC reports (xml.gz attachments to incoming emails) from different email providers.
I’ve added some tips about first configuration on README
If you have a ChatGPT Plus account you can setup it without requiring an API key and is really straightforward and cheapest option at the moment. Use tha command line cli for configuration, it will guide you. The bare minimum is setting up a LLM provider than you can access the CLI and chat with your bot. Settin up also a messaging platform like Telegram is really helpful (also easy and free).
Mattermost integration works fine too, but is a bit more complex.
I wish there were a way to get a consolidated analysis of the various DMARC reports (xml.gz attachments to incoming emails) from different email providers.
It can access emails and I think your bot should be able to set up this. Just ask him to do it
Thank you.
I don’t own any payed AI-account (merely misusing a Perplexity Pro plan as a better search engine).
However, I have installed an Ollama instance on my MacBook Pro, and I have already considered installing one on my Proxmox as well.
Not really, you can use local models, but you need a pretty good one to do something useful.
BTW you can setup an OpenRouter account and pay per use.
Nous Research have their API portal that should give all the service that Hermes could use (web search, image generation, TTS, STT. The more services it can access to the more powerful it is.
I haven’t tried their portal yet, but I’ll do it soon as its a very good way to found their development.
I’d like to start with Ollama first and gain some initial experience. I’ve also configured the API key and Base-URL. But how do I specify an LLM modell available in Ollama?
At the moment, it’s still routing to Anthropic Claude?
API call failed (attempt 1/3): NotFoundError [HTTP 404]
🔌 Provider: ollama-cloud Model: anthropic/claude-opus-4.6
🌐 Endpoint: http://192.168.3.155:11434/v1
📝 Error: HTTP 404: model 'anthropic/claude-opus-4.6' not found
📋 Details: {'message': "model 'anthropic/claude-opus-4.6' not found", 'type': 'not_found_error', 'param': None, 'code': None}
⚠️ Non-retryable error (HTTP 404) — trying fallback...
❌ Non-retryable error (HTTP 404): HTTP 404: model 'anthropic/claude-opus-4.6' not found
❌ Non-retryable client error (HTTP 404). Aborting.
🔌 Provider: ollama-cloud Model: anthropic/claude-opus-4.6
🌐 Endpoint: http://192.168.3.155:11434/v1
💡 This type of error won't be fixed by retrying.
I think, cloning technology is almost here.
Can you clone yourself with AI. Your thoughts, philosophies, approach, styles, etc.
then put that Agent to Work.
We use Dokploy and coolify extensively to deploy websites, and web apps we build, for prod and demo.
We build, and commit to github
github actions, takes over, he site is deployed, and updated.
recently, Dokploy Added a Comprehensive MCP, which means, the deployment, is an end to end process, without Human in the Loop.
DNS, can now be written to file, Pushed to git, audited by AI, actions validate, Push Updates to DNS provider, Site is built with another agent, committed and pushed, Now we can create Research agents, for market an competitor analysis, Automated dashboard creations for ingestation
an agent that does social media posting (kilo uses one) and responds accordingly.
anaother that gets user feedback, and docuemnts it as issue
an agent that reviews the issues, does market analysis, assigns it to build agent, that implements, and the solution is built.
WHERE THE HECK ARE WE HEADING TO?
AGI/ASI was never a single Model
at the moment it’s quite hard, but for building ns8-hermes-agent i created a NS8_RESOURCE_MAP.md in the repo and in the AGENTS.md I told to “browse the authoritative docs, and gather similar code patterns or prior art”
Thats works a bit, but I’m not really satisfied.
I’m learning that by doing it and I’m refining my agentic engineering skills while doing it.
BTW we decided @ Nethesis to start including our agents and skills files in the repositories, I hope that the agentic engineering knoledge that we’re building compounds.
RN if you try to use an agentic harness to change ns8-hermes-agent it should know what to do as those files are in the repository and maintained. This repo is a kind of experiment for that.
It can do both, it has skills for using them. But I haven’t experimented yet to use another harness inside hermes
l:
