77,000 tokens consumed. 800 were useful.
— AlphaSignal AI (@AlphaSignalAI) 5 juin 2026
That's not a model problem. It's a filtering problem.
Headroom is an open-source repo sitting between your agent and the LLM.
It compresses tool outputs before they arrive. Original data is cached, retrievable in 1ms if needed.
Real… https://t.co/T8jkIrgtFn pic.twitter.com/lHob65JhaD
77,000 tokens consumed. 800 were useful. That's not a model problem. It's a filtering problem. Headroom is an open-source repo sitting between your agent and the LLM. It compresses tool outputs before they arrive. Original data is cached, retrievable in 1ms if needed. Real