Hi guys. New to Agent Zero but familiar with what I’d consider ‘normal’ token usage based on experience with n8n, bolt, etc. After spending the day with A0 (using requesty for LLM APIs), I’m constantly seeing single messages eating over 100k tokens across 10+ api calls - which gets huge very quickly. I’ve tried a lot of potential solutions - from adding the ‘true’ flag for caching in the LiteLLM input field - to trying to control api call volume and token usage via global behaviour instructions and going down rabbit holes trying to sort something out via Perplexity at the VM level. Is this standard with A0? That is - it’s just a token hungry monster that sends full system prompts, verbatim chat thread copy, summarised chat thread copy and more in every API call associated with every single message? I’m just trying to wrap my head around this because it just doesn’t make sense. And that could simply be down to the fact that I’m a ln A0 noob. Any guidance appreciated 👍

Hi folks, been staking A0T pretty much since the beginning to access Venice AI models, epic feature of A0! That said, for the first time I am seeing the multiplier not add up to the dynamic quota which at time of the screenshot was 3.4x but the dynamic quota is less than 2x Any idea what might be happening? Also @Jan Tomášek is there any plan to use more pool fees to add to the A0T pool and increase total compute available as more and more people join? I have noticed that as more people join, my balance continues falling which seems like a disadvantage to OG stakers who have been staking since the beginning I understand about locking up for a higher score and have locked up A0T for various periods even until 2028 as I have faith in the project and team plus use this compute daily to run many aspects of business and personal life Much respect just some honest questions from a long time A0 fan and staker, thanks

In the first three days of using Agent Zero, I burned over 400 million tokens (Opus 4.5, Gemini 3 Pro, Gemini 3 Flash). The problem of optimizing tokens with the agent was quite challenging, as each execution involved a large amount of memory and the steps weren't optimized. The fatal flaw was the tasks; when I scheduled tasks, they could easily fail, leading to infinite loops. The worst instance was on February 8th, 2026, when 200 million tokens were wiped out in just a few hours. In summary: - The final result is quite good, but the cost is too high. Even after setting up context length reduction and memory depth, 60-70% of tokens were saved, but the number of tokens burned remains quite large. Hopefully, future updates will address this issue.