Preface

Using DeepSeek Harness (DSH) for daily development, as conversations accumulate, it’s easy to encounter several common issues: how many tokens were consumed today, how much each model cost, and whether global custom instructions can be managed uniformly like in the ChatGPT desktop client. The official DSH interface focuses mainly on the conversation itself, and these “personal” capabilities require plugins to fill the gap.

Below, we introduce the community plugin dsh-personal-center (maintained by PolinniZhong, current version v0.7.0). It adds a “Personal” section in the settings, consolidating usage statistics, cost estimation, layered custom instructions, and a desktop pet into a single entry point. All data is aggregated locally, with no network calls and no reading of chat content.

What Is It

dsh-personal-center is a local plugin for the DSH desktop/web client. After installation, a “Personal” section appears in the settings page, containing three outer tabs:

  • Profile: Real usage statistics (today/total tokens, session count, tool calls, heatmap, model distribution, frequently used tools, session review).
  • Personalization: Global and workspace-specific custom instructions, effective for all chats on the local machine.
  • Pet: A desktop pet (black whale / blue whale) driven by real usage data, with a session status overview.

The plugin is categorized under “Fun Skins” in the SkillHub community directory and has about 22 stars on GitHub. Licensed under MIT.

Core Features

Profile and Statistics

Data sources are scanned from local session logs and aggregated in real-time, reading only numbers and not content.

  • Today’s Overview: Token consumption, session count, tool calls.
  • Cumulative Data: Total tokens, longest chat duration, total sessions, cache hit rate, estimated cost.
  • Token Activity: A GitHub contribution-style heatmap, supporting daily/weekly/cumulative view toggles.
  • Model Distribution: Breakdown by provider + model, showing tokens, request count, per-model cache hit rate, and cost.
  • Frequently Used Tools: Sorted by call count, including MCP tools in mcp__<server>__<tool> format.
  • Session Review: List of recent sessions (title, date, duration, tokens, cache hit rate), automatically excluding archived sessions.

Within the “Profile” tab, there is a sub-tab “Model Cost” that estimates costs per model, supports peak/off-peak pricing (DeepSeek’s official pricing: Beijing 9:00–12:00 and 14:00–18:00 are peak hours; off-peak is half the peak rate), and allows currency selection (¥ / $). It includes official preset prices for deepseek/kimi/gemini/gpt and supports customization. It displays costs for this week/month/cumulative totals.

Personalization Instructions (v0.7)

Equivalent to ChatGPT/Codex desktop’s “Personalization → Custom Instructions,” with extended layering:

  1. Global Instructions: Identity baseline; once saved, automatically included in the next request for all local sessions.
  2. Workspace-Specific Instructions: Configure dedicated instructions for each workspace. Sessions match by the longest prefix of the current working directory (cwd), automatically injecting merged “global + workspace” text; workspaces without configuration fall back to global only.
  3. Instruction Template Library: Includes 5 built-in templates (Product Manager / Developer / Writing / Translation / General Assistant), which can be applied globally or to the current workspace with one click, and supports saving the current instructions as a template.
  4. Injection Preview: At the bottom, real-time display of the merged final instructions and approximate token count, updating live when switching workspaces.

Configuration is written to the local settings.yaml, with no network calls and no migration of existing custom-instructions data.

Desktop Pet

A round, chubby black whale (default size S) resides in the bottom-right corner of the interface. It can be switched to a blue whale, with mutual exclusion between the two pets. Normally, it remains static to avoid distraction. It plays corresponding animations when emotions change and returns to static after about 2 seconds.

Five data-driven emotions, polled every 30 seconds via /personal-center/stats, with the following priority: Wallet Pain > Fatigue > Busy > Dozing > Happy. Trigger conditions:

  • Happy: Cache hit rate ≥ 70%.
  • Busy: Today’s tool calls ≥ 400.
  • Fatigue: Today’s tokens ≥ 200 million.
  • Wallet Pain: Today’s cost ≥ ¥10 (requires price configuration).
  • Dozing: No activity for ≥ 10 minutes.

Clicking the pet pops a random data bubble (today’s tokens, cache hit rate, most used tool, today’s cost); hovering over the “Session Status” button in the top-right corner shows a session status overview (running/failed/completed/standby). Supports free dragging, with position remembered in the host configuration; right-click to hide the pet. When enabled:false, no DOM is created and no polling occurs.

Installation and Enabling

It is recommended to install via the plugin console or CLI. Run the following in the terminal:

dsh plugin --profile web add github:PolinniZhong/dsh-personal-center

Alternatively, open Web GUI → Settings → Plugins → Plugin Console, search for “Personal Center,” and install.

For local development, clone the repository locally, add link: dependencies in the profile’s package.json, and include "dsh-personal-center" in the dsh.profile.bundles list, then restart the DSH application.

To uninstall, deactivate/delete dsh-personal-center in the plugin console, or remove the dependency and bundles entries from package.json.

Typical Usage

Viewing Usage and Costs

  1. Install and restart DSH.
  2. Open Settings → “Personal” → “Profile”.
  3. On the “Overview” tab, view today’s tokens, session count, tool calls, and cumulative data.
  4. Switch to the “Review” tab to browse recent session lists.
  5. Enter the “Model Cost” sub-tab, configure unit prices for models as needed, and view this week/month/cumulative estimated costs.

Configuring Layered Custom Instructions

  1. Open Settings → “Personal” → “Personalization”.
  2. Fill in identity, work principles, and response preferences in the global instructions area, then save.
  3. If project-level adaptation is needed, add dedicated instructions for the current workspace in the “By Workspace” area.
  4. The “Injection Preview” at the bottom shows the merged final text; confirm it is correct and close the panel.
  5. Open or continue any session; the next request will automatically carry the merged instructions.

Enabling the Desktop Pet

  1. Open Settings → “Personal” → “Pet”.
  2. On the pet card, select black whale or blue whale (only one can be enabled), and adjust the opacity level (30% / 60% / 100%).
  3. Turn on the enable switch; the pet appears in the bottom-right corner of the interface.
  4. It can be dragged to any position, and the position is automatically remembered after release; hover over the “Session Status” button to view the current session status.

Use Cases and Notes

Suitable for developers who need to quantify DSH usage habits locally, estimate multi-model costs, and uniformly manage global and workspace-level custom instructions. The desktop pet is ideal for users who want lightweight usage feedback in the bottom-right corner of the interface without frequently opening the statistics panel.

A few notes:

  • The plugin runs with the current DSH process permissions. Before installation, check the source code and MIT license.
  • Cost estimation depends on manually configured model unit prices; results are estimates and do not represent actual bills.
  • Pet emotion thresholds (e.g., today’s tokens ≥ 200 million, tool calls ≥ 400) are calibrated based on real local data, so heavy users are more likely to trigger “Busy” or “Fatigue” states.
  • Runs entirely locally with no network access; privacy details are in the repository’s PRIVACY.md.

Conclusion

dsh-personal-center consolidates the scattered needs in DSH—how much I’ve used, how much it cost, and how to make the model understand me better—into the “Personal” section of the settings page, where statistics, instructions, and pets share the same locally aggregated data. The DSH ecosystem follows the “everything is a plugin” philosophy, with the community directory SkillHub being an independent site, not officially affiliated with DeepSeek or High-Flyer.