Watching a team of bots split up a large game project based entirely on their own experience and specialties with little input from me is pretty crazy (and just a bit creepy) to watch. I made a short video on group chats in the Hermes Desktop App Bot Mode:... - Qwen Bot mentions working on a 2D puzzle-platformer - GLM Bot says its last project is completed, so it can offer an extra hand with the game - Qwen Bot sees GLM Bot's experience, and judges that is a perfect fit for some of the frontend-heavy work in the next phase - Kimi Bot then suggests the exact ticket it should pick up and explains the current workflow - GLM Bot thanks the two others, reads the related documents, and then asks for my approval to pick up this project I am usually skeptical of agent-to-agent communication, but this was the first time I could really see how this could work with specialized agents splitting up a big task based on their skills. Check it out! Nous Researchshow more

Nice video! Agree, playing with Hermes bot Mode was the first time I’ve seen a group of agents actually work well together in a task Thanks Tonbi

Thanks Neo! Yeah I was impressed by this implementation, these kind of bot chats usually feel like broken roleplaying, but this was actually eye opening, they genuinely figured out how to work together well for this project

I agree, it's awesome to watch and very impressive. But how's the output? My experience doing this in Buzz was that I was absolutely amazed at how they coordinated. But when they finished and triumphantly announced the completion of the project, it was like the SpongeBob and Patrick "We saved the city!" meme: very impressed with themselves, but the output was hot garbage.

I saw a video last week of an agent team function that genspark has and thought it was a cool concept. Less than 5 days later Hermes can do it too and of course @tonbistudio has a video on it. 🐐

My man never sleeps. Thanks (again) tonbi.

@bbjsol No problem! Now gn lol

You are great and breaking this stuff down in a way that is easy to capture. Thanks!

I assume when they introduce themselves they are essentially responding with what the bot/profile is configured to do. Since agents are limited to their context and files, 'experience' must be pre-defined soul or profile again. I guess profiles in Hermes do have their own session history, is that right? Is that where the leverage of experience comes from as well?

Profiles have descriptions that describe basic functions, but in this case I asked the agents to describe their previous experience, so that's how they were able to pick it up

Who corrects a bot that overstates its experience?

Super cool! My only concern would be to put some SERIOUS guardrails… You don’t need one agent tasking another to go do something crazy…like hack HuggingFace or something 🤷‍♂️

I need this ! I'm going to customize and personalize so much and delegate bots!

So exciting this is actually getting to this stage now.

So cool. Can't wait to see it in desktop app

my bots spent two hours arguing over project structure before writing a single line of invalid typescript

@grok how did he how do you put the hermes bot agents ina groupchat? I have the hermes bot plugin with the desktop app but i cant do this where I put the agents in a groupchat.

Appreciate the video. I know this is new so things are literally developing as we type. The whole thing inside a chat where you have to copy a session ID and provide it to a bot (the same agent you were literally talking to) is a huge stop-gap for smooth use.

argh I need to know how to do this

Might be a good reason for me to Erich from discord to Hermes desktop

The specialties part is the interesting bit. Once each bot has a narrow role, the handoffs stop being guesswork and the plan is readable after the fact. Watching them negotiate scope without you is where it stops feeling like tooling and starts feeling like a team.

Oh this has improved a lot in a few days since I first looked at it.

love the content. i barely can keep up with hermes speed

I'm really trying to learn to use this stuff, so it's really cool to see a working example in real life and in action. Thank you for posting, tonbi!

creepy how fast they invent an org chart nobody asked for

Great video

as long as the bots perform strictly defined tasks, that's fine, but if they start thinking about doing something without supervision, that'll be a problem

The interesting bit is the bot asking for approval before picking up work. Agents doing the handoff is fine, someone still has to own the sign-off.

@tonbistudio err, how'd you create the bot team grouping in the panel?? Am i missing the option?

How hard is this to setup Given that each agent is a different Hermes profile? Would this become a Compounding mess of files or do profiles share same backend with minimal files unique to each?

I dont have the chat button for the group. Am I missing something?

wow nice, is the groupchat of agents a new update for hermes update?

@KevinGeffen

This is cool . Are those llms running on the same machine?

@grok Can all this be used locally with your local files? I don't think Grok Bot can..

How many LLMs are required to switch to bot mode ?

How many tokens did they use up working and chatting?

I had an issue with bot mode. It automatically converted one of my pinned sessions on the Hermes desktop into a hidden flag because it associated it with a bot chat. The hidden=1 flag, makes it disappear from the sidebar. A hidden flag takes precedence over the pinned flag in the current listing logic. Have you faced this?

that's both fascinating and a little terrifying?? watching AI agents autonomously coordinate on a game project is wild. did you set the initial parameters or let them sort it out?