Hermes Agent, LLMs and SLMs

[littleman]

You all should be looking into Hermes.  A couple of years ago we had a conversation about AI driven projects that can manage other AI systems and get actual work done.  I am just diving in now, but it looks quite capable.

https://hermes-agent.org/

https://hermes-agent.nousresearch.com/

[rcjordan]

> I am just diving in now
> Hold my beer, Bubba, I'm going in

Installing on a laptop? linux mini-pc?  Your raspberry-pi?

Keep us posted.

[ergophobe]

> "give it your messaging accounts"

Mmm... I'm not comfortable with that. How are you using it and how are you sandboxing it?

[littleman]

>How are you using it and how are you sandboxing it?

Use Docker, a VM or a machine without any sensitive information.

https://hermes-agent.nousresearch.com/docs/user-guide/docker
https://openclawlaunch.com/guides/hermes-agent-docker

[littleman]

> "give it your messaging accounts"

I messed with email.  Google has a app token system where you can give an app a unique password and limited access.

[ergophobe]

So, for example, you can give it read only access to your email?

I guess the problem for me is that being mostly retired, I just don't have that much email volume and so I don't think feeding it into an LLM would be a big advantage for me.

[littleman]

I don't know about read access.  That might be a setting available in Gmail, but I haven't looked into it.  I did use it to send myself a test email.  When reaching out to the world I am going to have Hermes have it's own accounts so that there is a wall between it and my personal information.

I am setting up a dedicated "server", an old desktop I use to use for work.  I am going to be running it headless (no GUI interface) and use tmux to have a persistent session with Hermes that I can log in and out of while keeping Hermes running in the background. 

My goal is to build a low effort information based businesses with a subscription model.  I am still working out the plan, but it seems like an interesting experiment.  I want it to be 99% automated and I am going to document the process. 

It isn't going to be anything fancy, I will be using Hermes the way someone would be using a smart intern to get the bulk of the work done.

[littleman]

I have been doing a bit of digging on different models to try out. 

A couple of obvious ones are:

• site audit
• reputation management

Both would be fairly easy to deploy, with site audit being particularly easy and low cost.  The reputation monitoring/management would be a little harder because of unauthorized scraping  or having to pay to use someone else's scraped data.  What's turning me off to both of these is that they both have a lot of market saturation.  Though there are an endless amount of potential customers there is also an *almost* endless amount of companies trying to get their attention.

I have considered going ultra-niche and focusing on California school bus operators.  This is a space I currently now a lot about.  I would monitor the California Highway Patrol, California Department of Education, DMV, CARB, NHTSA, and the California Legislature and post any relevant/actionable news for the local industry.  The strengths here is that I am in that space and also that there isn't much being done in it.  The drawback is that news is slow, the market is small and there is potential liability if I and the bots get something wrong.  I think the liability part could be mitigated with disclaimers and with sighting links.  The potential customer base is also a lot of managers who work in public sector fleets, so there also may be obstetrical for them to subscribe to the service.

Two I am currently considering:

1: Broadening out the school bus monitoring service, but make it for all commercial vehicles. That makes the market 100 times larger and the potential audience is mostly private sector where spending for information that can save headache and money is a lot easier.

2: Scraping local government websites for municipal construction opportunities.  The upside here is pretty good.  Government websites are all over the place in the way they post information.  As far as I can see there is low liability if I give wrong information.  If the contractor misses a job posting that's bad for business, but I am not going to get sued over it.  It should be an easy purchase without much red tape.  I could start locally and expand -- repeating the model as I go.  Every city, county and school district posts bids and there are thousands of contractors (HVAC, electrical, fencing, paving, tree trimming, ect.).

[Rupert]

I am impressed, and as I read, my mind was going the same way. The difficult bit from my perspective is reaching the customers. What are you thinking, offering a free service at a basic level, and subscription for more?

Also, are your customers likely to share your tool? In the construction one I would guess they will be in competition, so are unlikely to want to share it; with the commercial vehicle one, they are more likely to be in touch and share a good resource.

  I am not sure I fully understand the plan for that one, but if drivers/managers can share useful tools that helps I would think.

I think what you are doing is as usual beyond me!  Its awesome!

[littleman]

For the information based ideas there really isn't any tools that I am seeing, more like notifications email/SMS.  For commercial fleets there is a hodgepodge of regulations in the United States.  We have the federal, state and local laws to navigate. 

On the construction side I am thinking to just shoot them an email offering them a free month.  If they find the information useful then they could subscribe.  I think this could actually be a valuable service.  I've been going through my local government websites (city/county/schools) and they are all over the place when it comes to layout.  I am sure it's quite a pain for the typical contractor to navigate all that.  I am thinking weekly summaries and alerts when a new project goes live.

[ergophobe]

>> commercial vehicles

Are you friends with bakedjake?

Last he visited, which is at least 5 years ago, he was doing something where they were gathering a bunch of data and somehow using that to help truckers find insurance. Something along those lines. Before that, he was doing some mapping thing with big data sets using a lot of Amazon Turk workers.*

> typical contractor

Have you asked local contractors? I just spent an hour talking to the local excavating contractor. I can ask how he finds out about county contracts.

I know that here there are two types of contracts. This is my terminology, but roughly they are

 - emergency work: pipe breaks, need someone ASAP, no bid. For that you have to be on the county list. I think most of that work gets awarded as a "push" - the county staff get on the phone and hire the first available.

 - regular work: publicly posted contracts that go out for bid to all takers


*Turk. Interesting aside - he said everyone was using Turk wrong because they are looking for cheap. He was looking for good. So he was paying people $25-30/hr and using it to find really good people who were not in the regular workforce because they, say, only had a couple hours here and there between classes or while the baby was napping. So they were $50/hr workers who were being offered $10/hr by everyone else. His strategy meant they earned double and he paid half. I don't know where that kind of arbitrage is possible now, but that sort of thing could help with homogenizing scraped data at a reasonable price.

[littleman]

I don't really know Bakedjake.  That's an interesting idea about hiring Turks and it makes sense.  I have been doing a lot of digging on the world of contractors.  There are several different types in California and they do different types of business.

From my initial research there are contractors at different levels.  The ones I do not want to target are the large "Class A" engineering contractors, "Class B" general contractors or the residential only "B-2".  What I want to target is Specialty Contractors "Class C". 

This could be broken down into different categories:
C-10 Electrical
C-20 HVAC
C-27 Landscaping
C-36 Plumbing
C-39 Roofing
C-8 Concrete
C-9 Drywall
...


The big players (Class A & B) subscribe to big platforms like "ConstructionConnect" and other "enterprise" level services, but those cost $5k to $10k a year.  The residential contractors wouldn't be interested in this service because it doesn't apply to them.

Class C contractors either work commercial, home owners or public sector. What I am aiming for is a service to make getting those public sector jobs easier to find.  The more I look into it, the more of a mess I find.  A contractor could manually have to check 50 plus sites to find work.  The good thing from my perspective is that California make these Class C contractors who are looking for public sector bids register and make their names available as public record.