Every token ashlr keeps out of context, counted.
ashlr trims tool output before it ever reaches the model — a measured −57% across TypeScript, Python, and Rust reference repos. The community ledger opens today; every synced session adds to the running total. Here’s exactly what it counts.
The running total starts at zero and climbs with every developer who opts in. Be the first on the board with /ashlr-leaderboard on. The numbers it accrues are exactly the kind below — real, measured, per-session.
What it’s counting, on a real codebase.
Every entry on the ledger is a byte-measured delta like these — what a tool would have returned versus what ashlr actually sent the model.
Reproduce on your own repo: bun run bench · full methodology →
Top savers
opt in with/ashlr-leaderboard onNo one’s on the board yet. Be the first — run /ashlr-leaderboard on and your GitHub handle + savings appear here after your next sync. (Opt-in is off by default; only your handle and totals are ever shown — never your email or code.)
- +Real savings, not estimates of intent. Every token figure is the byte delta between what a tool would have returned and what ashlr actually sent to the model, divided by the tokenizer ratio — the same math behind /ashlr-savings.
- +Deduped per developer. We sum the highest lifetime counter per user (a developer syncing from three machines is counted once), so the total never double-counts.
- +Dollars at a published rate. Tokens are converted at the Sonnet input price of $3 per million — the same rate the savings badge uses.
- +It's a floor, by design. Only developers who sync their stats appear here, and only aggregate counts ever leave a machine — no code, no prompts, no identity. The real total saved is higher.
Updated Wed, 17 Jun 2026 23:48:00 GMT · refreshes every 5 minutes.
Install ashlr and your savings start counting toward the community total — and your own /ashlr-savings ledger.