I just got the Hermes AI agent running on a raspberry pi five. Now trying to figure out what I want to do with it. Anybody running this and have any cool ideas?
@jfayre All kinds of random stuff. Sometimes my Sonos Move falls off the charging base; I have Hermes run a script every 30 minutes that finds the Move, and notifies me if the battery is under 10 percent and it's not charging. I have a hermes skill that uses the API for my weatherstation to give me the readings instead of using the inaccessible app. I have it check environment Canada every 30 minutes and notify me about any weather warnings. I have it monitor the load on my servers and let me know if CPU/disc/bandwidth get out of expected ranges. I have it run a weekly price check on groceries. I similarly have it run price checks on Amazon for products I'm interested in. The nice thing is that once Hermes writes the scripts and sets up the jobs, they don't cost any AI tokens to actually run.
@fastfinge Very cool! Did you use Hermes itself to write the scripts? What are you using to actually talk to it? I’ve been playing with the command line but it’s not the most useable thing in the world.
@jfayre I'm using nano-gpt.com. They have a $12 a month subscription that gives you 60000000 tokens a week to use with any open-source model. If you go to nano-gpt.com, click models, then filter from "all" to "subscription" you can see what the subscription includes. It's stuff like the Kimi models, GLM, deepseek, etc. So not the smartest, but 99 percent of the time I don't need a super smart model to answer a random question I have.
@fastfinge Hi! A few more questions. You said you are using Hindsight. Are you using their cloud service? Looks like local setup is a bit complex and requires I run anotehr hindsight service. Also, I'm having trouble iwth Basilisk throwing errors with some Hermes responses. Any suggestions? Thanks for all your help!
@jfayre I'm using it locally. When you set it up in Hermes, you can set it up as local embedded. Then it will set up postgresql and the rest for you, and you just need to give it an API key via an environment variable in your .env file. What error is basiliskLLM throwing? It may be an error with the model perhaps? Do you have Hermes set up as provider: openai, and does your custom endpoint URL end in v1?
@fastfinge The error isn't for every response. It seems to be when Hermes is offering me choices. Error is "error during completion: Object None is not subscriptable"
@jfayre You can set what toolsets are enabled per platform. Create a toolset with all tools but clarify, then enable only that toolset for the openai gateway.
@fastfinge@jfayre I do like Hermes, however I do sometimes have it help me with processes that are long running on SSH, E.G. installing/setting up stuff. Why the 30 seconds max process?
@Orinks@jfayre Because I don't want it installing something that can stay running on the server, and I want it to not run long-lasting commands. I will run those on its behalf, as they usually require a lot of resources and my server is busy, so I want to schedule them overnight.
@jfayre I did. Hermes writes the scripts and sets up the scheduled jobs to run them. I've currently got Hermes connected to XMPP, IRC, Discord, deltachat, and email. You can also have it create an OpenAI compatible API, so you can use apps like perspective inteligence or basiliskLLM with it. There are a couple of IOS apps that specialize in interacting with Hermes, but I don't love any of them, yet. Also, connecting it to a chat app is the best way to let it notify you. You can connect it to iMessage or whatsapp, but you need a second phone number for those, so I never did.
@jfayre On a more worrying note, as I look through its folders to jog my memory of everything it does, I've found that it's also created itself sets of scripts to control my hot tub, GE Oven, and instant pot. I've never asked it to do anything with those, or even mentioned that they exist. So that's...a thing that happened, I guess. When I die in a massive housefire everyone can blame Hermes. There's nothing quite like realizing that thanks to crappy device security and everything accessible insisting on being a smart home device, my AI agent could just murder me if it wanted to. I thought I had it isolated from that network, actually, so I'm not entirely sure what's going on here. But verifying Hermes can't turn on my oven or instantpot has suddenly become my top priority...
