Preface

DeepSeek Harness (hereinafter referred to as dsh) is an intelligent agent runtime open-sourced by DeepSeek AI, with the architectural slogan “Everything is a Plugin”. Models, tools, conversations, sandboxes and interfaces can all be split into plugins for combination. When using it daily for coding and running tasks, the usage has been increasing: how many Tokens are consumed per round, how high the cache hit rate is, and which workspace uses the most quota are not centrally displayed in the default interface.

Most people have gotten used to using the 53-week green grid style GitHub contribution graph to judge “whether I have worked this week”. The community plugin dsh-usage-stats maintained by Make0209 adds similar active grids, plus Token consumption, cache hit rate, account balance and workspace aliases, and puts them into the settings page of the dsh Web frontend.

Two points need to be clarified first. First, the DeepSeek Harness Plugin Directory is an independent community site and has no official affiliation with DeepSeek / HyperMind. Second, the official repository is still marked as developer preview, and the interfaces may be incompatible. This article is cross-checked based on the directory details page opened on August 18, 2026, the GitHub repository README and source code, and does not fabricate usage feedback.

What it is

dsh-usage-stats is a usage dashboard plugin for the DeepSeek Harness Web frontend, categorized under “Tools and Capabilities”, licensed under MIT, mainly written in JavaScript, with version number 1.0.0 in package.json. It is maintained by Make0209. As of August 18, 2026, the GitHub repository has 17 stars.

One-sentence positioning: Scan the locally persisted conversation logs, aggregate the number of rounds and Token consumption by registered workspaces, draw a 53-week GitHub-style heatmap in the settings panel, and add balance query and workspace alias functions.

There are other warehouses with the same name on GitHub, with different functions and installation commands. This article only covers Make0209/dsh-usage-stats. Please use the full repository path when installing, do not search only by the plugin name.

Core Features

The capabilities matching between the repository README and the source code can be divided into the following sections:

  1. 53-week usage heatmap. The grid color is close to GitHub green, and the count is based on “one light up per completed round”, including sub-agent conversations. The colors are divided into five levels: 0 times, 1 time or more, 3 times or more, 6 times or more, and 10 times or more. When hovering the mouse over a day, the number of rounds and input/cache hit/output Tokens for that day will be listed by workspace. Clicking the workspace chip can filter the heatmap and detail table.

  2. Statistical cards. The page displays by default: total Token consumption (input, cache hit, output, reasoning sub-items), cache hit rate, account balance, Token consumption progress bar for each workspace (the top 3 workspaces with the highest usage are listed first), total number of uses, and consecutive usage days. The cache hit rate is calculated as cacheRead / (input + cacheRead). The card numbers have a scrolling animation when first loaded, and the interface follows the light/dark theme of the Web UI.

  3. Time range. The upper right corner can switch between “Last 30 Days”, “Last 90 Days” and “All”, with the default being Last 90 Days. The heatmap itself covers approximately 53 weeks; the 30/90 day range mainly affects the card summary and workspace detail table.

  4. Workspace aliases. There is a “✎ Workspace Aliases” button in the header. Aliases are written into the KV unit usage-stats-aliases in $DSH_HOME/storages, and will remain after uninstallation or restart. The maximum length of a single alias is 80 characters; press Enter to save an item, and clear the input box to restore the folder name. You can also save all aliases at once.

  5. Data source. The number of uses and Tokens all come from the dsh persisted conversation logs. The Host half reads turn/end and assistant/message.usage, and listens to session/event for real-time folding. The plugin will backfill history when activated, and these logs will not be lost after uninstallation or restart. Only conversations that can match the registered workspace by the session cwd will be counted, and unmatched conversations will not appear on the graph.

  6. Account balance. The balance queries from DeepSeek Open Platform’s https://api.deepseek.com/user/balance, reusing the API Key configuration of llm-deepseek, and the plugin does not store a separate secret key. When the Key is not configured, the card will display a guide copy. The query result is cached by default for 5 minutes; the “Refresh” button on the page will carry force=1 to force a recheck.

The plugin package declares the dsh.bundle manifest and the Web client half, with dsh.client.platform set to web in package.json. The Host half is in lib/index.js, and the Client half is in lib/client.js, which registers a page named “Usage Stats” through the settings panel slot settings.section. The README states that there are no third-party dependencies in the package: the Host half only uses Cordis services, and the Client half only uses React provided by the module table.

Installation and Activation

The installation command given on the directory page is as follows, run it in the DeepSeek Harness terminal. The dsh CLI will parse the plugin from GitHub and install it into the current configuration:

dsh plugin add github:Make0209/dsh-usage-stats

This is a community plugin that will run with the permissions of the current dsh process, and may execute code during installation. Check the repository source code and MIT license before installing. For reproducible installations, pin the commit hash. The latest commit on the repository’s main branch on August 14, 2026 is 8992d306cdca8857b4362868d591fde3689765b0:

dsh plugin add github:Make0209/dsh-usage-stats#8992d306cdca8857b4362868d591fde3689765b0

The repository README also provides a method for Web profiles, which is consistent with the plugin’s platform: web declaration. If the current default profile is not web, you can use this command:

dsh plugin --profile web add dsh-usage-stats

The README notes that after installation, you only need to refresh the page, no need to manually modify the configuration or restart dsh. For local directory debugging, the README also provides manual registration steps: create a symbolic link (junction for Windows) under $DSH_HOME/profiles/node_modules/, then insert the following content into $DSH_HOME/profiles/web/cordis.patch.yml:

- insert:
    - id: usage-stats
      name: dsh-usage-stats

The user patch layer will hot reload, just refresh the page after saving.

Typical Usage

After installing and refreshing the Web UI, you can view the data in the following order. These steps correspond to the interface in the README and lib/client.js, and are not an extra manual.

  1. Open the settings and enter “Usage Stats”. The Client half mounts the page in the settings.section slot, and the tab is exactly these four words. The first time you enter, it may show “Loading usage stats…”, followed by a historical backfill progress bar with text similar to “Counting historical conversations scanned / total”. The progress bar will disappear after the scan is completed.

  2. View the cards and heatmap. If there are no conversations that can be attributed to a workspace on the local machine, the page will prompt “No usage records yet. Start a conversation and it will light up here.” When there is data, there will be six cards at the top, followed by a 53-week grid and workspace detail table at the bottom. The detail table lists rounds, input, cache hit, output, reasoning, total, hit rate and proportion.

  3. Switch time ranges. Click “Last 30 Days”, “Last 90 Days” or “All”. The heatmap grid will still be arranged in 53 weeks; the Tokens, hit rate, number of rounds on the cards, and the detail table will be recalculated according to the selected window.

  4. Hover and filter. Move the mouse over a grid to see the number of times and Tokens for each workspace on that day. Click the workspace chip above the heatmap, or click a row in the detail table, to only view that workspace; click again to cancel the filter.

  5. Edit workspace aliases. Click “✎ Workspace Aliases”, fill in the project name for the registered workspace, press Enter to save a single item, or click “Save All”. Aliases will appear on the chips, hover prompts and detail table titles, while the underlying path remains unchanged.

  6. View balance. When the DeepSeek API Key is configured in dsh, the card will display the currency and amount (the source code formats according to CNY / USD). If the Key is not configured, it will not crash with an error, but the balance card will be in a missing-key state. Clicking “Refresh” will force a balance recheck and refresh the statistical snapshot at the same time.

The data routes exposed by the Host half can also be used to understand where the page fetches data from, and are generally not called manually:
- GET /api/usage-stats: Statistical snapshot (including scan progress, daily data, workspace summary, aliases)
- GET /api/usage-stats/balance?force=1: Account balance
- POST /api/usage-stats/alias: Set workspace aliases

Applicable Scenarios and Notes

This plugin is suitable for people who are already using dsh web and want to clearly see their local conversation usage: to check whether the cache hit function is working, compare the Token proportion of multiple workspaces, or just want to use a yearly calendar to review whether they have been using dsh consistently recently. It does not support billing export, nor does it count conversations in unregistered workspaces.

Please note the following points before use, all from the README or source code, not inferred:

  1. Running permissions. The plugin runs with the permissions of the current dsh process, and may execute code during installation. Check the source repository and MIT license before installing.

  2. Only covers sessions that match workspaces. The statistics match the registered workspace paths based on the session cwd. Conversions with empty cwd or that do not match any workspace will not appear in the heatmap or details.

  3. Web UI only. package.json declares the client platform as web, and the settings page is only mounted on the Web frontend. Pure terminal / TUI usage is not within the scope of this plugin.

  4. Environment dependencies for balance query. The usage statistics itself reads local logs and does not require internet access. The balance query needs an API Key, and the Host half will call curl.exe or powershell.exe to request the official interface. The current implementation is clearly Windows-oriented; in environments without these two commands, the heatmap and Token cards are still available, but the balance card may always fail.

  5. Do not install the wrong plugin with the same name. There are at least other dsh-usage-stats repositories in the community, some doing multi-vendor balance queries, some doing CSV exports. The installation command on the directory page is github:Make0209/dsh-usage-stats, please use this one as the standard.

  6. dsh is still iterating rapidly. The official README states it is a developer preview, and destructive changes may occur. Pinning the commit is easier to reproduce than always tracking main.

Summary

dsh-usage-stats brings the 53-week grid style GitHub contribution graph into the DeepSeek Harness settings page, and adds Token breakdown, cache hit rate, consecutive usage days, workspace aliases and (when the Key is available) official balance queries. The data comes from local conversation logs, and will backfill historical data after installation, without building a separate ledger.

It is an MIT-licensed community plugin maintained by Make0209, not an entry in the DeepSeek official app store. The directory page and repository addresses are as follows. It is recommended to check the source code before executing the installation command:
- Directory page: https://deepseek-harness-plugin.com/zh-CN/plugins/dsh-usage-stats-make0209/
- GitHub: https://github.com/Make0209/dsh-usage-stats
- DeepSeek Harness official repository: https://github.com/deepseek-ai/deepseek-harness