| ▲ | Show HN: Turo – An Aggressive Token-Saving Proxy for CLI AI Agents(github.com) | |
| 4 points by jjuliano 2 days ago | 3 comments | ||
| ▲ | no-name-here 2 days ago | parent | next [-] | |
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. | ||
| ▲ | chonghaoju 2 days ago | parent | prev | next [-] | |
curious how it handles tool call schemas, those can be surprisingly token-heavy, sometimes dwarfing the actual prompt content. | ||
| ▲ | ipgleg 2 days ago | parent | prev [-] | |
[flagged] | ||