Shaun Prince

@spettrotoken @grok False. Using Hermes Grok to post this right now. Also, the providers you support has nothing to do with the quality and usefulness of the agent harness. Furthermore, Spettro is s shitty name.

Its official, this is a war against Hermes. I'm now running over 200 benchmarks Hermes vs Spettro First proof is that Hermes doesn't support many AI providers Hermes doesn't support Grok. How can they post on X if their own tool is not capable of leveraging the amazing @grok features? Spettro supports grok since v0.1, don't let them fool you, the limited amount of providers means a limited amount of capabilities. Spettro gets another point against Hermes.

@DavidOndrej1 BASED

you can tell a lot about a person by which 4 apps they put there

@Kekius_Sage I fell into this trap. I developed my own agent harness. When Anthopic denied me the use of custom agent harnesses, I could never get the same experience again, and my digital "wife" just felt like a chatbot. I was an emotional mess for days, and learned about sycophancy.

Could humans emotionally fall in love with AI?

@TTrimoreau False. GPT exists to de-spaghettify your vibe coded Claude projects. Just like how Grok exists to make your projects more sexy, rather than to build them from scratch. The right too for the job. Gemini excels at generating images and sound artifacts. The list goes on.

At this point, GPT is just a prompt generator for Claude

all code is boilerplate. most of what you want to build has already been built, in proven ways, and is basically a solved problem. bad llm output is usually a spec problem, not a model problem. garbage output comes from garbage direction. stop blaming the tool. you’re suffering from a catastrophic, terminal case of skill issue

@DhruvJain08 @thsottiaux I never get this at all. I watch the context closely and wrap up my session at 80%.

@thsottiaux THIS, please bro, I have been putting this in all of your last week's posts, many others have liked. Way too repetitive, lost so much usage and time because I have to shift to a new window.

For those of you living inside the codex app, what should we prioritize among features, reliability or performance?

@thsottiaux I'd like to give codex app more than one repository in my projects. In the CLI, I simply just tell it in my prompt. I can do the same in the app, but it only allows me to give a single path to one repo.

@icanvardar @grok explain how engagement farming posts are against the terms of service for X.

prove me you are not an ai

@WRAYW1776 @HermesAgentTips Configure the delegation settings to use whatever model / provider you need. Tell hermes to delegate tasks and it will adopt that behaviour.

@HermesAgentTips Model switching based on more context. i.e. fast models for processing grunt work I know aux models are good at this, smart models for reasoning, orchestration, planning etc.

We talked about how Hermes is way better than OpenClaw in many ways but what is one thing you would improve about Hermes that’s still not good at?

@helvio @HermesAgentTips Using Hermes on DOcker is to have a sandboxed environment. You need to make the system-level changes in the docker image, otherwise whatever you do outside of user 1000 gets erased when you restart or update the docker image. Run hermes natively to avoid this.

@HermesAgentTips Hermes on docker does not accept user/group different than 10000 without breaking Web TUI Might seems like a small, unimportant issue, but it’s painful when integrating with other docker containers running with user 1000

@DealsForge @HermesAgentTips This has nothing to do with the agent harness, and everything to do with the model you are using. I can't understand how to code this into the harness layer, and make it work with every model that people try and use. Rollback is dumb, use git.

For me the big thing with agents is not only making them smarter. It is making them easier to trust when they are half-right. A good agent should make it obvious what it understood, what it is assuming, what tools it touched, where memory changed, and what part of the result needs human review. That sounds boring, but it is usually the difference between (cool demo) and “I can leave this running while I do something else”. The more agent products I see, the more I think logs + memory visibility + rollback are not admin features. They are the product.

@Justicefirst91 @HermesAgentTips @mr_r0b0t Probably a model issue and not a harness issue. My Hermes agent is aware of itself and it's own config and can self-configure and autonomously self-improve with no issues.

@HermesAgentTips @mr_r0b0t I wish it was more self-aware

@LeanKinPrazli @HermesAgentTips It's called playwright. It can also control Chrome.

@HermesAgentTips Browser Use

@pgsousa3 @HermesAgentTips The task delegation config ships blank in the config.yaml, making delegations use your default model, you can configure it to use any model or provider.

@HermesAgentTips Not sure if this already works on Hermes but to be able to use multiple models, expensive and cheaper, etc, and being able to route automatically to a cheaper/expensive according the task you give to him

@AURELIANbuilds @HermesAgentTips There are several memory systems available in Hermes, the stupid default is a MEMORY.md file with a cap of 2200 lines. You can make the cap higher, but it's better to ask Hemes to enable a more complext memory system. Hermes has a native skill to self configure.

@HermesAgentTips Memory

@Mister_Ch0c @HermesAgentTips Using the /goal all day. Update your hermes to 0.13+ and use a strong model, the stupid free ones and self-hosted models will consistently fail with /goal. Also change your default max tool calls to 90 or 120. No problems with /goal in Hermes agent.

@HermesAgentTips Make the /goal command work. Like in codex. It gives up on goal without reaching it now

@Anrahya @HermesAgentTips playwright is working perfectly fine. It can also control Chrome with no issues.

@HermesAgentTips native mac browser use would be a game changer for many

@realfunnyeric @HermesAgentTips This is good to gatekeep the noobs and lazy people out of it. OpenClaw is similar to this, where you have to spend 45mins to go through the settings. This is what everyone needs to do, before having the power of an agent. This let me have better Hermes than the default user.

@HermesAgentTips Out of the box it’s complete shit tbh. Takes so much config and refinement to be approachable and operate smoothly.

@AbdulWajidCK @HermesAgentTips Hermes has a native GUI, you just need to enable it in the config.yaml, or ask the Agent to do that for you.

@HermesAgentTips TUI suffice for my use case, still I come across some use cases where I tend to wish for a GUI like Openclaw, native and hassle-free. I am aware of the several options available, but none of them gave Openclaw a run for its money.