FRESH

Hacker News

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

4 points by jjuliano

by no-name-here

0 subcomment

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

by chonghaoju

0 subcomment

curious how it handles tool call schemas, those can be surprisingly token-heavy, sometimes dwarfing the actual prompt content.

by ipgleg

0 subcomment

[flagged]