Hi Product Hunt 👋 I'm Adrien and we built capsul . It started with a boring problem: I kept running out of my Claude plan mid-task, on a Wednesday, every week. Not because I was working more, but because every turn of a long session re-sends everything that came before it. The question is a rounding error. The context around it is the bill. capsul sits between you and the agent CLI you already use, and sends it only what the task needs. Your existing Claude or ChatGPT subscription is what answers, so there is no API key to paste and no second model bill. The part I care about most is the benchmark. Every tool in this space claims 10x. So I ran 9 Claude models, 5 tasks each, 2 arms, 180 cells, discarded none, and published every interval, including the three models that lose an answer check with capsul. Median 58% less input, range 32% to 80%. When a row establishes nothing, it stays in the table anyway. Raw data and protocol: https://www.capsul.chat/benchmark Free for 10 prompts on your own repo, so you can measure it on your code instead of trusting mine. Happy to answer anything, especially on the methodology. What would you want measured that I have not?

capsul

Your Claude Code and Codex plan, stretched much further

Your Claude Code and Codex plan, stretched much further