dsh-cost-tracker

Using DeepSeek Harness for daily inference and development can quickly lead to a fog around your spending—especially when peak/off-peak pricing, cache-hit versus cache-write billing, and subscription quota tiers like Kimi Coding Plan all mix together. dsh-cost-tracker automates the bookkeeping: every API call is recorded with input, output, cache-hit, and cache-write tokens tracked separately; peak hours (Mon–Fri 9:00–12:00 and 14:00–18:00 Beijing time) are billed at full rate, off-peak and weekends at half, and local models such as Ollama are counted at zero.

The settings page ships a full visual dashboard: six summary cards (today / this month / total cost, API calls, tokens, balance), a consumption bar chart segmentable by peak/off-peak or by model, a 26-week token-usage heatmap, and per-model request and token-composition charts, all with hover-to-inspect daily details. A persistent time-band indicator at the bottom of the sidebar shows the current peak/off-peak phase and a countdown to the next switch; you can configure pop-up alerts and browser notifications before transitions. A status bar under the chat input splits this session's cost by the models and subscriptions actually used, collapsing to the top-two by default.

It is built for developers who make heavy daily use of DeepSeek models while also running a subscription plan such as Kimi Coding Plan: you can see the true out-of-pocket cost for on-demand calls and estimate the equivalent on-demand cost of subscription usage to gauge value. All data is persisted locally—detailed records are kept for 180 days, then auto-compressed into permanent daily rollups—so history survives restarts, uninstall leaves no trace, and one-click CSV export lets you take the numbers into Excel or Numbers for deeper analysis.