github alexgreensh/token-optimizer v5.11.3
v5.11.3 — Keep-Warm cache automation + dashboard light mode

5 hours ago

Keep-Warm (opt-in)

Stop paying the cache re-write tax on resumed sessions. Keep-warm sends a tiny pre-expiry cache-read ping (~0.1x) so your prompt-cache prefix survives pauses instead of re-writing at 1.25-2x on resume.

  • Opt-in, consent-first: off by default; one-time plain-language ask with a savings projection computed from your own session history
  • API billing only: subscription accounts are never pinged (request-gated quotas make pings net-negative there) — and never nagged
  • Honest books: savings are booked only when a resume verifiably lands on a kept-warm prefix; ping spend is always shown beside them, net first
  • Self-policing: a rolling tripwire demotes the feature automatically if it ever stops paying for itself
  • Probe-first policy validated by replaying 30 days of real history through the shipping code
  • Mechanism verified live: full-prefix cache hits, transcripts untouched, TTL refresh proven past the expiry window

keepwarm-enable to opt in (macOS scheduler installs automatically) · keepwarm-report for the receipts · keepwarm-disable any time

Dashboard Light Mode

Same visual language, light background: persisted toggle + prefers-color-scheme, WCAG 2.1 AA contrast on every themed pair (scripted audit, both themes), status colors re-derived for light backgrounds. Dark remains the default, pixel-identical.

Platform notes

Keep-warm ships for Claude Code (macOS scheduler; Linux/Windows: wire keepwarm-tick to your own timer). Codex: not applicable by design — OpenAI caching is automatic with no write premium, so there's nothing to recover. Other platforms: watchdog-only for now. Light mode: main dashboard + Codex; VS Code inherits your editor theme natively.

Don't miss a new token-optimizer release

NewReleases is sending notifications on new releases.