AI Agent Hub
Back to plugins
🖥️

dsh-peak

Client Updated 2026.08.24

Run the following command in DeepSeek Harness:

dsh plugin install geokkjer/dsh-peak

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install geokkjer/dsh-peak in your terminal to install; source is available at https://github.com/geokkjer/dsh-peak

About this plugin

DeepSeek moved its API to peak/off-peak pricing in mid-August 2026, and the peak windows line up almost exactly with Chinese working hours. The same long job costs twice as much at 3 p.m. as it does at 2 a.m. dsh-peak closes that small but costly information gap right at your input row, so you can see the current rate window before you hit run.\n\nOn the browser side, a tiny pill mounts on the left of the composer: a red PEAK badge during peak windows, a green OFF-PEAK badge otherwise, plus a live countdown to the next boundary and a hover tooltip showing both UTC and Beijing time. It self-refreshes every 30 seconds and flips colour at each window change, so you never need to do calendar math. On the host side, the agent can call the dsh_peak_status tool before proposing or launching expensive work, giving you a clear heads-up that you would be paying the peak rate instead of a surprise line item on the bill.\n\nIt is aimed at developers and teams running long generations, batch inference, or agent pipelines through the DeepSeek API who want to stay cost-aware without bolting on a separate dashboard. Zero runtime dependencies, plain ESM on both sides, and it activates the moment your web profile restarts.

Use Cases

  • Check the current rate window before kicking off a long generation
  • Schedule batch inference jobs into off-peak hours to cut the bill in half
  • Let the agent warn you about peak-rate costs before launching an expensive job

Best For

  • Developers running long generations or agent pipelines over the DeepSeek API
  • Engineering teams doing batch inference and tracking API spend
  • Individual users who want rate-window awareness without a separate dashboard