Well yesterday I reached a milestone, I have been building so much with AI that I burned through a million tokens. I honestly don’t know if that’s a lot or not, I just know that it caused my hermes agent to stop doing stuff because it didn’t have any more tokens to do things.
In just a matter of days, I have had my agent do a bunch of cool (and secret) stuff that I won’t get into here because it involves investments, research and other money related activities.
The good news is that I stumbled upon some tools that will help improve token use streamlining. One example is DEFUDDLE. This “skill” essentially discards all the advertising waste on web pages and gets to the good stuff. When you tell an agent to fetch something from the Internet, it has to read through all the junk just like you and I do unless you have an ad-blocker. DEFUDDLE is essentially an ad-blocker for AI. It not only removes annoying ads, it saves tokens from being wasted. I discovered this after I burned through a million tokens. You would hope that something like this should be built in to all these tools to save on electricity too but I guess that requires too much forward thinking and this stuff is all too new.
I had to pay more money for more tokens and I am now contemplating things like OpenRouter.ai to manage token efficiency and spread my AI stuff across the three paid subscriptions I am using now: Claude, Gemini, OpenAI.
Yes, I am now paying for three AI service providers and quite frankly they are worth the money….for now. It’s clear to me though that I need to be very pragmatic about what I build and spend money/tokens. AI makes too many things too easy but once you have to pay for those things, you start thinking about the value received.
My hermes agent is absolutely amazing and if Apple ever gets its own AI to do what hermes can do on their next phone, I may finally upgrade my iPhone 13. Too bad Apple seems lost in the clouds (pun intended).
Share The Wealth
Have you been punished for burning a million tokens yet?
What model(s) do you use with Hermes? I’m trying both Nanoclaw and Hermes and they are awesome. However, I’m using Claude Opus 4.6 with Nanoclaw and deepseek-v4-flash with Hermes. Deepseek is literally 50-100x cheaper than Claude, so I use my Nanoclaw AI bot only for complex stuff, and Hermes for simpler things. It’s like Nanoclaw is my scientist/engineer and Hermes is my secretary/personal assistant! I know any LLM model can be used with Hermes and I could have a “smart” Hermes as well, but I enjoy playing with different AI’s. Deepseek is rather abrupt and actually rude sometimes! Claude seems like a smart scientist/engineer. Gemini likes to find excuses for not doing what you ask until you insist… they really do have their own personalities.
On Hermes, I am using OpenAI for now. I initially just set this up to test but the more I test it the more I am hooked. I burn through a lot of Gemini credit because I do most of my app development on Antigravity but I recently started playing with Claude and I am preferring it to Antigravity however the token use on Claude seem far more stingy.
I am going to continue to test but I may switch to OpenRouter.AI for models and test the cost efficiency there soon. Now that I’m unemployed, I have all the time in the world!