dsh-token-api

Working long sessions in the DeepSeek Harness Web UI means constantly wondering how many tokens have been burned, how much context-window headroom is left, whether the highest reasoning level is actually active, and roughly what the round is costing—all of this lives in RPC responses and console logs with no at-a-glance surface. dsh-token-api injects a compact floating panel in the bottom-right corner of the Web UI that consolidates input, output, cache-read, and cache-write token counts, context-window pressure, the active reasoning level, and a cost estimate into a single glance, eliminating the need to dig through logs or fire off ad-hoc RPC queries.

The panel goes beyond a read-only dashboard. Reasoning level is rendered as a draggable slider that switches between off, high, and max by calling the official session.selectModel RPC under the hood. For cost estimation, the Host side scrapes the DeepSeek official pricing page every 30 minutes and relays the data to the browser via a same-origin route, automatically applying different unit prices for peak hours (Beijing time 09:00–12:00 and 14:00–18:00) versus off-peak hours. If the fetch fails, it falls back to bundled default prices; users can also manually override any model's price, and the override persists across reloads via localStorage. Styling uses Harness official --dsw-alias-* design tokens so the panel blends seamlessly into both light and dark themes. Data auto-refreshes every four seconds, and everything runs purely client-side: no model input is modified and no API key is ever read.

If you spend extended sessions in the DeepSeek Harness Web UI running long coding or deep-reasoning tasks and want an always-visible, real-time token-budget and cost gauge without writing a separate monitoring script or scraping pipeline, this plugin fills exactly that gap in the interface layer.