AI Agent Hub
Back to plugins
🖥️

dsh-token-hud

Client Updated 2026.09.07

Run the following command in DeepSeek Harness:

dsh plugin install chy007-fun/dsh-token-hud

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

Run dsh plugin install chy007-fun/dsh-token-hud in the DeepSeek Harness terminal to install the plugin; the source code is available at https://github.com/chy007-fun/dsh-token-hud .

About this plugin

DSH's built-in stats line has a blind spot: it only shows tok/s after the provider reports usage and the step finishes. Free-gateway models frequently omit usage or hang mid-stream, leaving you unable to tell whether the model is generating slowly or has already died. dsh-token-hud estimates output speed directly from stream chunks (roughly 1 Chinese character per token, ~4 characters per English token), so the counter moves the moment text appears without any provider report. When usage does arrive via assistant/message, the estimate is automatically corrected to the precise value, eliminating the initial ~10% error.

The plugin renders as a draggable, collapsible overlay in the bottom-right corner of the DSH web page, aggregating every session into a single view. A five-color status light makes state instantly readable: pulsing green for active output, yellow for waiting on the model, pulsing orange during a retry loop, blue for tool calls, and grey for idle or offline. Each model is listed with cumulative output tokens and peak rolling-window tps, both held in host memory and cleared naturally on restart with zero persistence. Position and fold state are remembered via localStorage; polling pauses when the tab is hidden; five consecutive API misses display a grey offline pill.

It is well suited for DSH users on free-gateway or third-party models with unreliable usage reporting; anyone running multiple sessions or agent streams simultaneously who wants a single aggregated speed dashboard; and those pairing dsh-llm-retry who need a visual cue of orange flicker during retries and green on successful output to confirm link health at a glance.

Use Cases

  • Estimate output speed from stream chunks when free-gateway models omit usage reports
  • Get a single aggregated overview of per-model speed and status across concurrent sessions
  • Pair with dsh-llm-retry to confirm link health via orange retry flicker and green output

Best For

  • DSH users on free-gateway or third-party models with unreliable usage reporting
  • Users managing multiple concurrent sessions or agent streams who need a unified speed dashboard
  • Developers who need visual confirmation of model link health and retry status