Cache Watch
ExperimentalSee prompt-cache usage and estimated time left in each thread. Inspect cache reads and writes above the composer, or enable time-limited keepalive from the Cache Watch page. Keepalive sends ordinary paid agent turns through your configured provider; cache hits and savings are not guaranteed. Requires BB 0.48+ and Plugin SDK 0.6.104+.
Overview
View sourceSee the latest reported prompt-cache reads, writes, and estimated time left directly above each thread's composer. Expand the right-aligned Cache summary for exact counts and the report timestamp. The context-card layout includes the same data in a Prompt cache section.
Cache estimates
The countdown updates locally every second. Green means the estimated deadline is in the future; red means it has passed. These are estimates, not a live cache status check. For Codex and Central Codex, missing expiry uses a visibly marked 30-minute assumption when cache activity was reported. Other missing telemetry stays unknown. Reports can aggregate several model requests in one turn.
Optional keepalive
Use the Cache Watch page or bb cache-watch to enable paid keepalive for a
selected thread and a limited duration. No watch starts automatically.
Keepalive sends ordinary visible turns asking the agent to reply OK. It can
cost more than recreating a cache and cannot guarantee a cache hit or savings.
Active threads are skipped; stop cancels future and deferred plugin pings.
Requirements
Requires BB 0.48 or later with Plugin SDK 0.6.104 or later and an existing configured agent provider. No separate API key is needed. Provider access and ordinary agent usage remain subject to that provider's billing or allowance. This plugin has no native silent-prewarm integration.
More in Token Usage & Limits 32
Understand or control token, context-window, and provider-quota use.



