dsh-token-rush

You hand the model a long task, stay absorbed in the conversation, and only wince when the invoice appears and your balance has quietly shrunk. Worse, DeepSeek doubles its per-token rate during weekday peak windows (09:00-12:00 and 14:00-18:00 Beijing time), and most users have no idea they are being charged twice.

dsh-token-rush pins a compact ~216 px floating HUD to the top-left corner of the DeepSeek Harness session page so the answer is always in your peripheral vision. It shows the current session cumulative token usage (input, output, and cache buckets) in real time, estimates cost against the official per-model price sheet with automatic 2x peak markup (marked with an approximation sign), and proxies the account balance query on the host side with a 60-second cache so the API key never leaves the server. Before any message is sent the window is fully transparent and click-through; once a conversation starts it fades to semi-transparency over 0.8 s. Drag it anywhere you like-its position is remembered locally, and a double-click snaps it back to default. When a peak window is active the entire HUD shifts to a bold amber-red, a gentle nudge that says: double pricing is on, so take it easy.

It is built for developers and individual users who call the DeepSeek API through the DSH frontend on a pay-per-token plan and want glanceable cost awareness without hunting through billing dashboards. Pure hand-written JavaScript, zero build steps, zero extra dependencies, MIT licensed-install it and it just works.