Show HN: Claude-thermos – keeps your Claude session warm for

Show HN: Claude-thermos – keeps your Claude session warm for you

What happened

claude-thermos is a tool that keeps Claude's prompt cache warm to prevent cache expiration during long sessions with subagents.
Cache rebuilds accounted for approximately 22% of the total bill across roughly 185 measured local sessions.
Individual cache collapses re-write between 200K to 500K tokens.
When served from cache, prompt cache costs 0.1x the input price.
Re-encoding at write rate costs 1.25x the input price.
claude-thermos launches Claude Code behind a local reverse proxy that points ANTHROPIC_BASE_URL at a loopback port.
Warm requests cost a cache read at 0.1x while prevented rewrites would cost a write at 1.25x on a much larger prefix.

Cache rebuilds accounted for approximately 22% of the total bill across roughly 185 measured local sessions. (22)
Individual cache collapses re-write between 200K to 500K tokens. (200000)
When served from cache, prompt cache costs 0.1x the input price. (0.1)
Re-encoding at write rate costs 1.25x the input price. (1.25)