dsh-cache-billing

The DSH context ring tells you how many tokens you used but never how much it cost. A high cache hit rate is great, but the longer your context grows, the more expensive the cached prefix becomes each turn. Near the window-switch threshold, every action in the old window is paying for cache you may never see itemized. dsh-cache-billing puts that bill front and center.

At the bottom of the overlay, three blocks—current step, current turn, and session total—stack top to bottom. Each shows a headline row with the grand total and token count, then indented lines for cache hit, cache miss, and output. Peak and off-peak rates are applied automatically based on event timestamps: weekday peak hours charge full price, everything else half, independent of your system timezone. V4 Flash, V4 Pro, and Vision Exp each have distinct unit prices, calculated per step by the model actually used. Third-party relays are billed the same way—no DeepSeek official route required. If the model name matches the rate table, an estimated amount is shown.

Ideal for two profiles: heavy DSH users constantly probing the window edge who need a real-time number to decide when to switch, and users connecting through relays like SiliconFlow whose pricing differs from official rates but who still want to see actual per-turn costs. Install, restart the DSH web, and it is live—no config file edits needed.