AI Agent Hub
Back to plugins
dsh-bottom-stats preview

dsh-bottom-stats

Client Updated 2026.08.25

Run the following command in DeepSeek Harness:

dsh plugin install 318197375/dsh-bottom-stats

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

Run dsh plugin install 318197375/dsh-bottom-stats in DeepSeek Harness to install this plugin (source: https://github.com/318197375/dsh-bottom-stats), then restart the web profile to activate it.

About this plugin

The stock stats line in the DSH web UI has a subtle but frustrating flaw: once a conversation grows long enough and input tokens climb into the hundreds of thousands or millions, the trailing output-tokens group is quietly swallowed by an ellipsis. dsh-bottom-stats exists to fix exactly that. It adds two persistent rows beneath the composer so you keep a complete, at-a-glance picture of performance and usage no matter how far the dialogue stretches.

The first row is a re-arranged stats strip covering turns/steps, LLM and tool latency, time-to-first-token and throughput, cache-hit rate, and input/output tokens. It spans and centers on the full main-column width, so nothing gets clipped regardless of content length. The second row is a context-occupancy progress bar, segmented into three colour-coded zones for system prompt, tool definitions, and conversation messages, with the occupancy percentage and used/total tokens on the right (e.g. 18% 177K/1M). Hovering the bar reveals the full reading and a per-part breakdown. All figures are pulled from the same sessionStats, tokenUsage, contextPressure, and contextBreakdown projections the stock UI uses, and labels follow the active UI language automatically.

If you run long multi-turn sessions in the DSH web interface and want to stay on top of token burn rate and context-window pressure in real time, this plugin slots in after a single restart with zero configuration — pure-JS distribution, no build scripts, and no extra authorization required.

Screenshots

Use Cases

  • Tracking full token burn and per-stage latency across long conversations
  • Inspecting context-window breakdown by system prompt, tools, and messages as the window fills
  • Comparing LLM versus tool invocation latency side by side
  • Monitoring TTFT and throughput trends over multi-turn sessions

Best For

  • Developers running long multi-turn sessions in the DSH web UI
  • Users who need real-time visibility into context-window pressure
  • Anyone who wants a complete, never-truncated stats readout at a glance