DeepSeek-Harness-chat-billing

Running DeepSeek models at scale inside DeepSeek Harness turns cost into a black box: your balance lives in an API doc, per-turn spend requires mental math, and a single day across many sessions is impossible to eyeball. DeepSeek-Harness-chat-billing brings that visibility straight into the web session header. A two-line badge shows your real remaining balance and this conversation's cumulative spend, while the detail panel breaks expenses down by model into cache-hit input, cache-mash input, and output tokens, capped by a today session-spend ranking so you can see at a glance which conversations consumed the most.

Pricing is not a rough guess. Every message is priced against DeepSeek's official peak and off-peak rate card (weekdays 09:00-12:00 and 14:00-18:00 Beijing time are peak; weekends are always off-peak), across the three billed token buckets. Session spend increments locally with each new reply and needs no extra network call; the balance is fetched only on mount, session switch, manual refresh, or page reload, so there is no background polling. Each completed turn also stamps a static cost amount at the end of the actions row, and turns with zero DeepSeek usage or failed loads simply stay hidden rather than fabricating a zero.

Built for developers who run multiple DeepSeek V4 sessions in parallel and want live, per-model cost transparency without leaving the chat. Configure a DeepSeek API key, install inside any dsh profile, and the billing overlay is ready with no additional runtime dependencies.