dsh-usage-dashboard

Long-running, multi-model, multi-session workloads in DeepSeek Harness tend to leave cost as a blur: where the tokens went, which model is the most expensive, and how much the peak/off-peak schedule actually saved are questions that normally require digging through raw logs by hand. dsh-usage-dashboard solves this by mounting a compact peak/off-peak billing dock directly below the dsh web input box. A single click expands into a full analytics dashboard that aggregates scattered cost data into an interactive, filterable visual panel.\n\nThree pillars carry the plugin. First, real-time peak/off-peak pricing: weekday peaks (09:00–12:00 and 14:00–18:00 Beijing time) and off-peak slots (half price, with weekends billed entirely at off-peak) are computed automatically, and the dock shows the current time-slot badge, countdown to the next price change, cumulative session cost, and account balance refreshed every 60 seconds via the official balance proxy. Second, cross-session and cross-restart persistence: every step tokens, cost, model, and peak/off-peak tag are written to a local JSONL file with a 40 000-line soft cap that trims the oldest entries. Third, multi-dimensional analysis with per-day drill-down: clicking any day on the cost-trend line reveals a 24-hour hourly bar chart, a same-day session leaderboard, a token-structure donut, and a model-distribution bar, all linked to global filters (time range, session scope, model) that re-render every card and chart instantly without a round-trip.\n\nThe plugin is aimed at developers and teams who run extended DeepSeek API workloads, juggle several models or concurrent sessions, and want peak/off-peak scheduling and cache-hit savings folded into day-to-day cost tracking. A DeepSeek API Key configured once in the model settings page is all that is needed for balance and pricing data to refresh automatically; no extra backend service is required.