Recall — Claude Usage & Context Tracker

Claude forgets — and you can't see it happening. Recall makes it visible. As a conversation fills the context window, Claude quietly drops earlier details: answers get shorter, instructions get missed, quality slips. Claude.ai gives you almost no warning before that happens — or before you hit your 5-hour and weekly limits. Recall is the missing dashboard. ━━ WHY RECALL, IF CLAUDE ALREADY SHOWS SOME USAGE ━━ Claude's own UI shows a bare percentage, buried in Settings, only after you go looking. Recall puts a live overlay right where you chat and adds the things Claude never tells you: how close this chat is to "context collapse," whether your next long prompt will get cut off, and exactly when your limits reset. ━━ WHAT YOU SEE, LIVE ━━ ▸ Context window % — how full this conversation is, updated as you type, with a clear "collapse likely — ~N messages left" warning ▸ 5-hour window & weekly cap — your real quota use with a countdown to reset ▸ Per-message token badges — every turn shows its token cost and share of your budget ▸ Truncation warning — flags prompts likely to be cut off and offers to split them into steps before you waste a message ▸ Output-size prediction — see Short / Medium / Long before you send ▸ One-click hand-off — auto-summarize a maxed-out chat and continue in a fresh one without losing the thread ▸ Streaming status bar — live input committed and output-so-far during generation ━━ 100% PRIVATE, LOCAL-FIRST ━━ ▸ No servers. Nothing you type is ever uploaded. ▸ No account, no API key, no signup. ▸ Your usage numbers come from the same endpoint Claude's own Settings page uses, read with your existing browser session. ▸ All data stays in your browser. Optional anonymous stats are off by default. ━━ BUILT FOR ━━ Free, Pro, and Max users — heavy Claude users, engineers, writers, and anyone who has watched a long chat quietly get worse and wondered why. ━━ REQUIREMENTS ━━ Chrome 88+ (Manifest V3). Activates only on claude.ai — nowhere else.