Hermes Agent - AI agent

[gpuerrilla]

Its very new to me I am a rookie with PCs just interested in AI. What are you using it for?

Learning and setting up locally. I have tried most of them to see each curve (ollama, lm studio, WSL, llama.cpp - windows) and extended that on a Frankenstein rig running ubuntu. I use a pi for searxng web searches and the models tap into that via MCP.

 

[jpod]

Hi is the MCP on the cloud?

Can Claude desktop do MCP?

Searxng is a search engine with emphasis on privacy...

 

[aaronyuri]

Hi is the MCP on the cloud?

Can Claude desktop do MCP?

Searxng is a search engine with emphasis on privacy...

The MCP server is an ancillary service/component to something. MCP is the protocol and then an MCP server will listen for MCP requests from clients. Where it's hosted can vary - there's nothing to stop you knocking together a basic MCP server using a Python library (like FastMCP) and self hosting it.

An LLM will have a set of tools/skills that it can use. One of those tools/skills can be an MCP client to interact with an MCP server. It's the new standard for programmatically integrating models with data and systems.

Claude has MCP capabilities, yes.

 

[gpuerrilla]

Hi is the MCP on the cloud?
...Searxng is a search engine with emphasis on privacy...

I used one of my pi's as I have my regular gaming/work machine and the dedicated ubuntu machine which can use it. This way I can update the one source and multiple machines can just point to this - rather than having to have my gaming machine on for example (the pi is low power and left online with any lightweight services).

I use the frontier services of Grok, Copilot, Gemini as they are free to a point, you can even have your hermes fallback to them but I tend to use mine local and private for work and research. You should also take a look at notebooklm for topics and reference purposes they are all great tools while you settle for what you want to do.

Months ago it was harder to get local models to have online search capability, as time has ticked over - most of this stuff is easier to setup now.

 

[jpod]

Thank you for sharing your expertise. I am a notebook llm fan already.. Just starting to learn to keep up a little

 

[Emlyn_Dewar]

I’ve used it for a while, and think it’s pretty great as a locally running service. Only ever hooked it in to Qwen 3.6 27B and Gemma 4 31B, but it’s interesting to chat to about various topics, and it’s even helped me improve my speeds on llama.cpp, in turn speeding it up.
Pointing it at a GitHub project and telling it that I have no idea how to implement it, it’s happy enough sorting things out even using these local models.
Don’t think I’d be happy using it on a machine that wasn’t purely dedicated to messing around though. Even with it asking permission before it does things.

I’m much just a hobby user, and don’t have much interest in frontier models. If I can’t download and run it locally, I don’t want to build a reliance on it. Qwen 3.5 397B-A17B was amusing to run, if only at Q4, but single digit tokens per second aren’t it…

 

[XeNoNF50]

I been running Hermes Agent since Feb. It is amazing how far it has come. It is a superb tool. I've personally got a Gemma 4 31B, a Qwen 3.6 27B, a Gemma 4 12B and a Qwen 3.5 35B A3B across an RTX 5090, RTX 4090, RTX 3080 and my Macbook Pro M5 Pro 48GB all running on my Hermes Agent. Alongside some cloud models. Tempted to get a DGX Spark now that Laguna S2.1 model has released which is basically perfect for it.

 

[p1RATE]

I'm so glad to see a thread on this now. I was scanning the forum for discussion on agentic tools like this and openclaw and was so surprised that a tech forum like ocuk had zero mention of any of it. I was actually using it as a barometer to see when agentic AI starts to get more widely adopted. I thought especially for how long the crypto section has been active and how the techy tinkerers should be the first to use it. It seems like it's just the 'entrepreneur bros' who have adopted it so far, they seem like the same archetype as the crypto bros from years ago

I'm using it pretty much every day for both work and personal tasks. It has been difficult to explain to people why it is so great, as the use cases are all so individual, but it is quite amazing what it can achieve.

I used it a lot for work, mainly lead generation and automating tasks. Stuff like monitoring my inbox giving me updated on what's active and drafting responses, posting to social media and automating video creation. I have some pretty cool reporting dashboards too. I've been able to cancel quite a few SaaS products and replace it with this.

For personal tasks it's great too. I asked it to list a bunch of furniture on Facebook marketplace. I just took a picture of the order sheet and said go do this. It created all the listings, descriptions and photos all by itself. Like a true Jarvis.

It can get a bit spendy though. I've given it tasks and before I realised it, it's racked up big charges. For example I gave it a task to pull data from Google maps, ended up with about £4k of charges from Google cloud services. I had another to look up details for keynote speakers, £1.5k of brave search API calls. Then running it on Claude (which is the best model for figuring out tasks and getting things done) it would burn about 20-50 a day in token usage. I use my GPT subscription for that now though and just have other models like opus and Kimi k for specific tasks to limit token spend.

 

[mid_gen]

I do think we could do with an AI subforum by now. There are so many threads on the topic scattered all over the forum it's hard to remember where the discussions are taking place.

Client of mine is using Hermes more and more. If there's an issue with one of the systems, he just asks Hermes in Slack what's wrong and it goes off and roots around AWS and our MCP gateway to figure out what's been going on. Can then raise a ticket and assign to one of the autonomous coding agents. The human-in-the-loop just reviews the PR and merges it.

 

[gpuerrilla]

I'm so glad to see a thread on this now. I was scanning the forum for discussion on agentic tools like this and openclaw and was so surprised that a tech forum like ocuk had zero mention of any of it. I was actually using it as a barometer to see when agentic AI starts to get more widely adopted. I thought especially for how long the crypto section has been active and how the techy tinkerers should be the first to use it. It seems like it's just the 'entrepreneur bros' who have adopted it so far, they seem like the same archetype as the crypto bros from years ago

I was one of them crypto bros but transitioned to LLM as they became possible to set up (about two years ago). Its much easier now to install and follow guides so adoption is massive - you can tell since we started 2026. For me the moment it was adopted so to speak was when all these mac mini's were flying off the stores selling out as people cottoned on that the RAMpocalyse was meaning an Apple product became good value!

I do think we could do with an AI subforum by now.

Totally agree, and like @p1RATE I was expecting more noise from this type of forum. We did have a folding forum which predated the crypto era but that has long sailed.

 

[XeNoNF50]

I do think we could do with an AI subforum by now. There are so many threads on the topic scattered all over the forum it's hard to remember where the discussions are taking place.

Client of mine is using Hermes more and more. If there's an issue with one of the systems, he just asks Hermes in Slack what's wrong and it goes off and roots around AWS and our MCP gateway to figure out what's been going on. Can then raise a ticket and assign to one of the autonomous coding agents. The human-in-the-loop just reviews the PR and merges it.

Yeah I think I would be somewhat more active here if there was. In reality there should be. It goes hand in hand. It is one of the main use cases of computing hardware nowadays. And it is not solely limited to just something like Hermes Agent / Openclaw, there are so many harnesses, so much software, so many local LLMs, so many cloud models now to tap into, both from closed source and open source labs in the West & China. Deserves it's own subforum rather than being fragmented and scattered across random subforums.

 

[p1RATE]

There is a thread for this

AI Section in forums

Shouldnt there be one. In a personal capacity im keen to understand how people use it. This is surely going to be transformative as people learn how useful it is. Im finding CHATGPT immensley useful...it has saved me time and money in practical ways. 1. PC Technical Fault Troublshooting. 2...

forums.overclockers.co.uk