MIT-licensed DSH plugin that turns the calls passing through `llm/stream` into a local billing ledger with per-platform / per-model breakdowns, peak-off-peak pricing, manual price imports, official-bill reconciliation and HTML / PDF / CSV export.
DSH integration
Native runtime
Author-claimed
Safety audit
Unaudited
Last verified
2026-11-13
License
MIT
01What can it help you accomplish?
See what DeepSeek Harness model calls actually cost, per platform and per model
A sidebar dashboard with calls, failures, aborts, input / output / cache-hit tokens and spend, drillable from a day down to hourly buckets
Developers running several providers or relay platforms through DSH who need to know where the money goes
Reconcile your local ledger against the provider bill
A per-platform, per-day comparison of the official amount (plus optional call count and token totals) against the plugin ledger, with the difference shown line by line
Anyone who has to explain or check a provider invoice before settling it
Hand a spend report to someone who does not run DSH
A self-contained HTML report or PDF document grouped by platform and model with time details, or CSV for spreadsheets
Team leads, finance and clients who need a file rather than a live dashboard
02How to install into DeepSeek Harness
Prerequisites
- DSH Web UI in the range `>=0.2.1-rc.5 <0.3.0` with Node.js `>=20` — the README verified Windows 11 with DSH `0.2.1-rc.5` and `0.2.2-rc.1`
- A DSH profile whose package.json you can edit (the README registers the plugin as a `link:` dependency)
- `pdfkit` is already a dependency and installs automatically; `@deepseek-ai/schemastery` is an optional peer
Installation steps
- 01
Put the plugin directory into your workspace
- 02
Register it as a dependency in the profile package.json — the `link:` form is enough
- 03
Restart the app; the sidebar then shows the billing statistics entry
- 04
Upgrading from 1.12.x: uninstall the old bundle in Plugin Manager first (old names `dsh-billing-insights`, earlier `llm-usage-insights`) so two bundles do not register the same sidebar page
Verify the integration
- The sidebar shows the billing statistics entry after the restart, and the page lists calls grouped by platform / model
Rollback
- The README documents no uninstall command: it only states that uninstalling does not delete the ledger, so delete `$DSH_HOME/llm-usage-insights/` by hand if you want the data gone
03DSH integration and capability boundaries
DSH plugin bundle registered as a `link:` dependency in the profile package.json — it adds a sidebar entry and a page that reads the DSH `llm/stream` usage records
Per-platform / per-model usage breakdown
the calls recorded by DSH `llm/stream`→calls, failures and aborts, input / output / cache-hit tokens and spend per platform and model, with hourly buckets under each day
appends counters, tokens and price sources to `usage.json` in `$DSH_HOME/llm-usage-insights/`Peak / off-peak and tiered pricing
token counts plus a price table→spend computed with peak windows (Beijing time by default, `utcOffsetHours` adjustable) and tiered prices, booked to 6 decimal places
writes a price-source marker per row so an estimated price is never shown as an official oneManual price import (table or CSV template)
unit prices typed into the platform dialog, or a filled-in CSV template→persisted manual prices, optionally with peak windows such as `09:00-12:00, 14:00-18:00` for every day / weekdays / weekends
manual prices survive catalogue refreshesan explicit save is required before a DeepSeek official price turns into a manual overrideOfficial bill reconciliation
the amount (optionally the call count and tokens) copied from the provider console per platform and date→a line-by-line comparison against the plugin ledger with the difference shown
never logs in to or scrapes the provider console — you paste the numbersReport export (HTML / PDF / CSV)
a chosen date range and whether to include time details→a self-contained HTML report, a PDF document, or CSV
PDF export needs `pdfkit` and a local Chinese font, otherwise it errors instead of emitting a garbled report
04Who is it for? When not to use it?
Good for
- Developers running several providers or relay platforms through DSH who need to know where the money goes
- Anyone who has to explain or check a provider invoice before settling it
- Team leads, finance and clients who need a file rather than a live dashboard
Not for
- Only calls that pass through DSH `llm/stream` are recorded — requests made while DSH is not running, direct API calls and other clients never enter the ledger, so the numbers are DSH-scoped, not account-scoped.
- Amounts are estimates computed from models.dev, the DeepSeek official price or your imported manual price; settlement is by the provider bill, and the reconciliation step only puts the official numbers next to the local ones.
- Data lives only on this machine under `$DSH_HOME/llm-usage-insights/` with no cloud sync or multi-machine merge; hourly detail is trimmed after 30 days (`retainDays`), monthly totals are kept, and uninstalling does not delete the directory.
05Compatibility, maintenance and safety notes
- Only calls that pass through DSH `llm/stream` are recorded — requests made while DSH is not running, direct API calls and other clients never enter the ledger, so the numbers are DSH-scoped, not account-scoped.
- Amounts are estimates computed from models.dev, the DeepSeek official price or your imported manual price; settlement is by the provider bill, and the reconciliation step only puts the official numbers next to the local ones.
- Data lives only on this machine under `$DSH_HOME/llm-usage-insights/` with no cloud sync or multi-machine merge; hourly detail is trimmed after 30 days (`retainDays`), monthly totals are kept, and uninstalling does not delete the directory.
- Built for the DSH Web UI in `>=0.2.1-rc.5 <0.3.0` on Node.js `>=20`; the page routes only accept same-origin / loopback access, and PDF export fails outright when no Chinese font is available.
MIT · README evidence only, latest release v1.14.0, safety not audited
06Frequently asked questions
How is the plugin installed into DeepSeek Harness?
Put the directory into your workspace, register it as a `link:` dependency in the profile package.json, then restart the app — the sidebar shows the billing statistics entry. The README targets the DSH Web UI in `>=0.2.1-rc.5 <0.3.0` on Node.js `>=20`. Coming from 1.12.x, uninstall the old `dsh-billing-insights` bundle first so two bundles do not register the same sidebar page.
What does it write into the DSH workspace, and does it keep my prompts?
It writes into `$DSH_HOME/llm-usage-insights/`: `usage.json` holds only counters, tokens and price sources, `manual-prices.json` holds your imported prices, and the models.dev catalogue is cached next to them. The README states no prompt or reply content is stored. There is no cloud sync, and uninstalling does not delete the directory.
Does it block or budget my DeepSeek Harness calls?
No. It never intercepts model requests: it only observes the calls that already pass through DSH `llm/stream`, and there are no daily / monthly budgets, overage blocking, balance queries or auto top-up — it only counts and displays.
How are prices decided when a platform is missing from models.dev?
The lookup order is custom `pricing` → manual import → DeepSeek official auto price (official API only) → models.dev catalogue → built-in reference price, and the page marks which source was used. Relay platforms billed per call can switch the dialog to pure per-call pricing, where token and peak prices no longer apply; `chargeFailedCalls` decides whether failed or aborted requests are charged too.
Why is my PDF export failing?
PDF export depends on `pdfkit` and a local Chinese font — without a usable font it errors instead of silently producing a garbled report. Use the self-contained HTML report instead, or install a Chinese font on the machine.
07Related DSH workflows
weknora
by tencent
Open-source LLM knowledge platform: turn raw documents into a queryable RAG, an autonomous reasoning agent, and a self-maintaining Wiki.
honcho
by plastic-labs
Memory library for building stateful agents
webnovel-writer
by lingfengqaq
A long-form Chinese web-novel writing assistant for Claude Code and DeepSeek Harness, designed to reduce memory loss and hallucinations in serialized projects of up to 2 million characters.
reme
by agentscope-ai
ReMe: Memory Management Kit for Agents - Remember Me, Refine Me.
08Data and sources
`dsh-billing-statistics` 是 DSH 的本地计费账本：把经过 `llm/stream` 的模型调用按平台、模型和小时记下来，用价格表估算花费，并提供报表与官方账单对照。它不接管模型请求，也不替你登录任何平台。
- 按 **平台 / 模型** 汇总调用次数、失败与中止次数、输入 / 输出 / 缓存命中 token、花费。
# 1. 把本目录放到工作区 # 2. 在 profile 的 package.json 里登记为依赖（link: 形式即可） # 3. 重启应用，在侧栏看到「计费统计」入口
This page is generated from the project’s public documentation, repository metadata and a structured parse of DSH Plugins; last verified on 2026-11-13. Found an error? Submit a correction.
Best DeepSeek Harness Plugins
Twelve plugins worth installing first — picked from the whole catalog, across every category.