HERMES AGENT: Как экономить до 80% токенов. Простой способ.

Today I'll tell you how to save up to 80% of your Hermes tokens. The setup takes 10 minutes, is done once and then it works automatically. Let's look at what tasks the agent performs in the background, which model to use for each of them, and configure them using the mouse in the interface. And then I'll

show the same thing in the terminal. It depends on what you're used to. At Hermes, the same operation on an expensive model and a cheap one can cost 10 times more. The quality will be the same. There's a gift at the end. file by which the agent will configure itself. Pick it up on

Telegram. My name is Maxim Skorokhod, and here everything is about neural networks and my projects. Let's go. So, let's take a look at what's happening right now in Hermesi. Most likely , you installed Hermes, chose the main model, eh, most likely something powerful: Opus or the latest GPT. This completes the setup. The logic is, in

principle, clear. I need a convenient smart agent, and I take the best model. Everyone advises this. Only the agent talks to you. Well, if it's in half of the requests. The rest of the time it does service work that you didn't order and don't see in the interface. Let's imagine. One working day.

You posted a screenshot to Telegram. The agent sent the image to the model to understand what was in it. They gave him a link to the documentation, he downloaded the entire page and asked the model to extract the essence from it. The dialogue grew and reached the ceiling of the context. The agent fed the entire

story into the model and asked it to compress the context. Sum, maybe you saw, wrote the command to delete, and it first found out if it was safe. They started a new session and he asked the model to come up with a name for it. Five appeals, of which you saw zero. And each one was sold at the rate of the most