Glyphh cuts Claude Code token burn

This reads like a serious retrieval layer, not a flashy AI wrapper: if the index is good, Claude needs fewer tokens because it starts with the right files and relationships.

• –The biggest win is not raw speed, it’s forcing the agent to query a purpose-built code index before it burns context on repo spelunking.
• –Commit-triggered reindexing and drift scoring could be genuinely useful for change-impact and release-risk signals if they stay calibrated.
• –The downside is familiar: bad tokenization, weak thresholds, or stale indexes will quietly turn “smart context” into confident misrouting.
• –The repo’s `glyphh_search`, `glyphh_related`, and `CLAUDE.md` workflow makes this feel like a developer-tooling layer, not just a research demo.
• –If the claimed savings hold at scale, this fits neatly into the emerging market for context infrastructure around AI coding agents.