Prompt caching vs the long LLM conversation: where your input bill actually hides

Short summary

PromptCrunch is a proxy that optimizes LLM conversation history before sending to the API, reducing input token costs for long sessions. It deduplicates superseded code, compacts stale tool output, summarizes old turns while preserving recent context, and achieved 75% input-token savings in non-cached windows and 7-10% in cached ones. Two-line setup; it pairs with prompt caching to address the full long-session cost lifecycle.

• •PromptCrunch optimizes conversation history to reduce API costs
• •Deduplicates code, compacts output, summarizes old turns
• •Delivers 75% token savings in non-cached windows; integrates with prompt caching

Generated with AI, which can make mistakes.