Preface

DeepSeek Harness (DSH) is a CLI tool designed for AI agent development, built around the core design principle of “everything is a plugin.” During real-world development and debugging, developers need to continuously track key metrics such as token consumption, cache hit rate, actual costs, and API latency to optimize model invocation strategies and control expenses. dsh-metrics-panel is a monitoring panel plugin designed for the DSH ecosystem. It provides real-time statistics in the web interface, addressing the pain points mentioned above.

Features

This plugin mainly provides the following capabilities:

  • Real-time Statistics: Tracks token usage, cache hits, costs, and latency.
  • Visualized Display: Provides overview trends, request details, and error logs.
  • Multi-dimensional Aggregation: Supports aggregating usage and costs by model, provider, or tool.
  • Billing Configuration: Supports configurable unit prices for calculating actual costs.
  • Interface Customization: Supports theme switching (light/dark/follow system) and panel color scheme switching (5 presets), and supports refreshing all historical data.

Installation and Activation

Before installing, make sure that the DeepSeek Harness CLI (@deepseek-ai/dsh) is installed. Install it using the official plugin command:

dsh plugin --profile web add github:bulai-z/dsh-metrics-panel

After installation, start the web interface to use it:

dsh web

Usage

After opening the page, find and click the “Monitoring Panel” entry at the bottom of the sidebar.

  • Time Filtering: Use the time range selector at the top (1h/24h/7d/30d/90d) to view data across different time dimensions.
  • Cost Configuration: Set the unit price and currency in the “Costs” section. After saving, the results are recalculated in real time.
  • Historical Data: Click “Refresh All History” to backfill data from sessions that have not been opened.

Notes

  • Data Persistence: Panel data is based on in-memory state during runtime. Data is cleared after the plugin stops or the process restarts.
  • Security Masking: The Authorization field in HTTP request examples is masked as <redacted>.
  • Manifest Check: If you see the message declares no dsh.bundle after installation, it means the dsh.bundle declaration is missing in package.json, and the plugin cannot be activated.

Conclusion

dsh-metrics-panel is a practical auxiliary tool for DeepSeek Harness, suitable for developers who need to monitor model invocation costs and performance in real time during development. For more details, you can view its GitHub repository or the community directory.