My Claude Code “second brain” is useful, but its distiller is burning an absurd amount of tokens

TiagoDCPP · reddit · 2026-08-30

The author built a persistent 'second brain' around Claude Code that distills session transcripts into a knowledge base. It's useful but consumes vast tokens (approx 2.28B logical tokens in a month). Analysis reveals waste from duplicated tool results, metadata, incorrect tool configurations, and re-distilling old sessions. A redesign is planned to optimize cost.

Original post →

More from coding & agent

coding & agent channel →