jkStars/dsh-token-usage-stats

English | 简体中文

DSH web plugin: cross-session token usage, request count, and optional cost analytics (ctx.tokenUsageStats), with a self-contained dashboard page at /token-usage-stats (JSON feed at /api/token-usage-stats) and a sidebar footer entry opening the dashboard in an in-page modal.

Screenshots

Sidebar footer entry — click 「用量统计」 to open the dashboard

Dashboard — today view (cost stack, token trend, breakdown, per-model and top-session tables)

Breakdown tooltip (hover a bar for per-category details)

All-range daily view

Note: Only DeepSeek official API costs are currently counted (token usage, request counts, and cost are replayed from local Harness session logs); it does not include API costs from other providers or channels.

Install

dsh plugin --profile web add dsh-token-usage-stats@0.3.15

The package's cordis.patch.yml inserts the plugin row; the browser half loads from the dsh.client manifest. Restart the host (or reload the GUI) after installing.

To update to a newer release:

dsh plugin --profile web add dsh-token-usage-stats@latest

Usage

Open the dashboard from the sidebar footer entry, or browse directly to http://<host>:<port>/token-usage-stats. The page defaults to today's hourly view and offers today / 3-day / 7-day / 30-day / all ranges, where the 7-day, 30-day, and all ranges show the daily view (the all-range view automatically downsamples to weekly or monthly intervals over longer spans to keep the charts readable and uncluttered). It auto-refreshes every 10 seconds. Cost figures appear only when model pricing is configured.

Config

The inserted row accepts config.currency (report cost in this currency) and config.pricing (per-model per-million-token prices). Cost is computed with a peak/off-peak split: peak hours are Beijing time 09:00-12:00 and 14:00-18:00, every other Beijing hour is off-peak; weekends (Beijing Saturday/Sunday) are always off-peak. A model priced with a peak/offpeak pair uses the matching tier by the usage record's time; a model priced with only the four flat keys uses that price at any hour. The default row ships currency: CNY and peak/off-peak pricing for deepseek-flash and deepseek-v4-pro (with historical model alias fallback).

Example override in the profile's own cordis.patch.yml:

- patch:
    - id: token-usage-stats
      config:
        currency: USD
        pricing:
          deepseek-flash:
            peak:
              uncachedInputPerMillion: 2.0
              cacheReadPerMillion: 0.04
              cacheWritePerMillion: 0
              outputPerMillion: 8.0
            offpeak:
              uncachedInputPerMillion: 1.0
              cacheReadPerMillion: 0.02
              cacheWritePerMillion: 0
              outputPerMillion: 4.0

Development

pnpm install   # devDependencies install the dsh 0.2.0-rc.2 @deepseek-ai/dsh-* types from npm
pnpm run build # tsc -> lib/types, tsdown -> lib/index.js + lib/client.js

The devDependencies are pinned to the same 0.2.0-rc.2 releases as the target runtime, so type checking runs against the exact API the host loads. peerDependencies declares ^0.2.0-rc.2, which the dsh host evaluates on install and boot (evaluatePluginCompatibility); an unsatisfied range refuses the installation or skips the bundle. Those peers are provided by the dsh host at runtime, not installed from npm.

Publish

npm run build
npm publish    # runs prepublishOnly (npm run build) automatically

Repository

• npm: https://www.npmjs.com/package/dsh-token-usage-stats
• Source: https://github.com/jkStars/dsh-token-usage-stats

Issues and pull requests are welcome.