Show HN: Claude-thermos – keeps your Claude session warm for you (github.com)

A new tool named Claude-thermos has been introduced to optimize cost management for users of the Claude Code AI model. This tool addresses a significant issue where the prompt cache of the main agent expires after five minutes of inactivity, especially when waiting on subagents, leading to costly re-encodings of conversation history. Claude-thermos maintains the cache "warm" by issuing minimal requests during idle periods, preventing the expiration of the cache and enabling users to avoid what can amount to about 20% of their processing costs. The technical implementation of Claude-thermos involves running it as a local reverse proxy that monitors traffic and manages cache sessions. It automatically triggers a warming request without disrupting ongoing user interactions, significantly reducing the costs associated with rewriting large amounts of token data. The tool logs session details, including the number of warm requests sent and savings achieved, allowing users to track financial benefits. By minimizing the need for full re-encodings, Claude-thermos stands to drastically improve cost efficiency for those utilizing long sessions with multiple subagents in their AI workflows.