AI Agent Hub
Back to plugins
dsh-token-dashboard preview

dsh-token-dashboard

Client Updated 2026.08.25

Run the following command in DeepSeek Harness:

dsh plugin install tsja2001/dsh-token-dashboard

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

Install this plugin in DeepSeek Harness by running dsh plugin install tsja2001/dsh-token-dashboard; the source code is hosted at https://github.com/tsja2001/dsh-token-dashboard

About this plugin

When working with DeepSeek Harness, token consumption is often a blur: how much of your billed input is cache hits, which step inflates TTFT, and how sessions compare side by side. dsh-token-dashboard brings all of these metrics into a single Statistics tab in DSH Web, making every token account transparent at a glance.

The plugin layers five views: a per-session token overview (billed input, output, cache hit rate, context usage), a token composition breakdown (uncached input, cache reads, cache writes, and output shares), a request-efficiency panel (rounds, steps, LLM and tool latency, average TTFT, decode throughput), a step-by-step drill-down (model, status, reasoning tokens, TTFT, tok/s), and a cross-session comparison with sorting that distinguishes current, running, and sub-agent sessions. All data is read directly from DSH pre-computed projections; the plugin has no server, database, telemetry, or network calls of any kind, making it a purely local, read-only display.

If you call DeepSeek Harness frequently in your daily workflow, want to pinpoint context-pressure sources, verify cache-utilization health, or compare usage across multiple sessions, this plugin is ready to use out of the box with zero configuration and zero extra token overhead.

Screenshots

Use Cases

  • Pinpoint where tokens are spent and how much hits the cache in a session
  • Diagnose inflated TTFT or low decode throughput in a specific step
  • Sort and compare token usage across sessions, including sub-agent overhead

Best For

  • Developers calling DeepSeek Harness who want full token-account visibility
  • AI engineers tracking cache efficiency and context pressure with quantified metrics
  • Teams comparing usage across sessions and debugging performance bottlenecks