Preface

The core philosophy of DeepSeek Harness (DSH) is “everything is a plugin.” When running long conversations or debugging complex prompts in DSH, it is difficult to assess context pressure or the marginal cost of a single operation without real-time visibility into Token consumption. The dsh-where-did-my-tokens-go plugin solves this pain point by mounting real-time badges next to the “Deep Thinking” indicator, allowing developers to intuitively see Token usage for input, output, and the thinking stage.

Plugin Overview

This is a DSH plugin maintained by community developer RevxZxCx under the MIT license. It is not an official DeepSeek product. It mainly solves two problems: how to view Token counts in real time within the conversation flow, and how to distinguish between Tokens consumed during the model’s “Deep Thinking” stage and Tokens consumed for the final answer.

Core Features

The plugin provides the following capabilities:

  1. Real-time input/output badge display
    Next to the “Deep Thinking” indicator, badges update dynamically to display Token counts. in N represents input Tokens, out N represents output Tokens, and think N represents Tokens in the output that belong to the thinking stage.
  2. Deep thinking/answer split display
    By using reasoningTokens, it splits output Tokens into a thinking portion and an answer portion. The thinking portion is a subset of the output rather than an additional component, meaning out = think + answer.
  3. Automatic unit conversion
    Supports automatic unit conversion for K (thousand), M (million), G (billion), and T (trillion) units, such as 1866 → 1.9K, and provides a unit-locking feature to preserve readability on specific sites.
  4. Merged/split display modes
    Provides both merged and split view modes, which can be switched in the settings.
  5. Per-action Token usage tracking
    Each time an action (such as “thinking completed”) finishes, that row displays use N token, showing the Token consumption of that step itself rather than the total consumption for the entire session.
  6. Per-site independent configuration
    Supports customization for specific sites, including font, color, and unit settings, and includes visual settings preview cards.

Installation and Enabling

Before installing, ensure your DeepSeek Harness version is 0.2.0-rc.2.

  1. Run the installation command
    Use the following command to add the plugin to the desktop profile:
    dsh plugin --profile desktop add dsh-where-did-my-tokens-go

If you need to install it from GitHub, you can use:

    dsh plugin --profile desktop add github:RevxZxCx/dsh-where-did-my-tokens-go
  1. Restart DSH
    After installation, you must restart DeepSeek Harness for the plugin to take effect.

Typical Usage

After installation and restart, use it in the following ways:

  • View real-time consumption: After sending a message, observe the badges next to the “Deep Thinking” indicator. They update in real time as streamed output is generated.
  • Distinguish thinking from answering: Keep the “thinking” toggle enabled and check the ratio between think N and out N to estimate how many Tokens the model spent on reasoning.
  • Switch display mode: Go to Settings → Token Counter and click the preview card to switch between merged and split modes.
  • View per-step cost: Check each “completed…” action row to confirm the specific Token consumption for that step.
  • Adjust styles: In the gear menu on the settings page, adjust the font, color, and unit for a specific site, then click “Save” to apply the changes.
  • Preview effects: The preview card on the settings page supports hover animations and real-time switching, allowing you to preview the merged and split effects without saving.

Notes

  • Version dependency: The plugin depends on @deepseek-ai/schemastery.
  • Data source: Input Tokens are determined by the harness’s prompt-size projection. Output Tokens are estimated during streaming (CJK characters approximately equal 1 Token, Latin characters approximately equal 4 Tokens) and are updated to authoritative values after streaming ends.
  • Security: Unsaved draft settings are not leaked into the real-time badges.

Conclusion

Through refined UI design, this plugin deeply integrates Token counting into the conversation flow, making it suitable for DSH users who need fine-grained control over costs and context. For more details and source code, visit the GitHub repository or the community catalog.