4 ms·

Show HN: Turo – An Aggressive Token-Saving Proxy for CLI AI Agents

• ipgleg 2mo ago
  [flagged]
• no-name-here 2mo ago
  The repo would be more helpful if it had, or linked to, any kind of quantified data on how this affects the LLM output quality or task completion. > Why > System prompts are 50-200k tokens. … 1. Are system prompts really the primary purpose for this? 2. As of five months ago, Claude’s prompt was 12-65k tokens depending on enabled tools [1]. Does “50-“ refer to 50K? The repo should link their source for the 50-200K range. [1] https://www.reddit.com/r/ClaudeAI/s/9OiIiHcIn3 https://www.reddit.com/r/ClaudeAI/s/9OiIiHcIn3
  • jjuliano 2mo ago
    [flagged]
• chonghaoju 2mo ago
  curious how it handles tool call schemas, those can be surprisingly token-heavy, sometimes dwarfing the actual prompt content.
  • jjuliano 2mo ago
    [flagged]