dsh-agent-pricing

Long coding sessions in DeepSeek Harness can turn into a cost black box. You keep prompting without a clear sense of how many tokens have been consumed or what the session is actually costing. dsh-agent-pricing closes that visibility gap by keeping a live cost readout pinned right under the composer, refreshed every five seconds with a CNY / USD toggle, so you always have a rough handle on your balance burn while you work.

Beyond the at-a-glance price tag, the plugin surfaces a daily-usage overlay that breaks down input, output, and cache tokens, renders an hourly bar chart, and shows a per-model cost table. Two registered agent tools, price_estimate and session_cost, let the model itself quote a per-call estimate or report full-session spend mid-task, so you no longer have to ask it manually each time.

Pricing data stays aligned with the official DeepSeek page: on activation and every 30 minutes the plugin scrapes current rates plus the announced peak and off-peak tiers (effective 2026-08-16 16:00 UTC) and matches each call to the correct tier by its own timestamp. A graceful fallback chain means a network hiccup never breaks estimation. It is ideal for anyone running the DeepSeek Harness Web profile who cares about token spend and wants the agent to self-report its burn rate. No API key is required, session logs stay on your machine, and the only outbound requests go to the public pricing page and the repo's raw price file.