插件目录 / Data / dsh-agent-pricing
dsh-agent-pricing
已验证 · 实测可装 Way2LOose4
功能简介
实时会话成本预览:费用、图表、工具、同步官网
可用 — 实测通过,早期项目
实时会话成本预览:费用、图表、工具、同步官网 实测能干净安装、正常启动。早期项目,但功能可用。
「已验证」表示我们的自动化 CI 在干净 profile 里实际执行了 dsh plugin add 并启动成功——仅此而已。功能描述与版本兼容性均为作者声明。这不是安全审计,也不代表对第三方代码的背书。
README
dsh-agent-pricing
Real-time session cost preview for DeepSeek Harness: a live cost readout under the composer, a today-usage overlay chart, and
price_estimate/session_costagent tools — with prices kept in sync with the DeepSeek official pricing page (peak / off-peak aware).
Features
- Live cost readout under the composer — the current session's accumulated cost, refreshed every 5 seconds, CNY ⇄ USD toggle. This is your approximate balance-consumption preview while you work.
- Today overlay — today's input/output/cache tokens, an hourly usage bar chart, and a per-model breakdown. The overlay also shows which price source produced the numbers, when the table was last verified, and — after peak/off-peak billing takes effect — whether the peak or off-peak tier applies right now.
- Agent tools —
price_estimate(per-call estimate) andsession_cost(full-session usage read from the same source as the Web stats strip), so your agent can quote its own burn rate mid-task. - Prices always aligned with the official site — on activation and every 30 minutes, the plugin scrapes the official DeepSeek pricing page: current rates plus announced peak/off-peak tiers with their effective date. Any failure degrades gracefully through the fallback chain; a network outage never breaks estimation.
workspace agent-pricing.json (1) your override — wins over everything
↓ absent
official pricing page scrape (2) DeepSeek rows, incl. announced peak/off-peak tiers
↓ down / HTML changed
this repo's prices.json (3) edit it on GitHub → every install re-aligns,
↓ down no reinstall needed
built-in table (4) offline fallback — ships with announced tiers
DeepSeek peak/off-peak billing (announced effective 16:00 UTC 2026-08-16, peak hours 01:00–04:00 & 06:00–10:00 UTC) is computed per usage event at its own timestamp: calls made before the switch stay at flat rates, peak-hour calls bill at peak rates automatically.
Installation
Requirements: DeepSeek Harness (Web profile), 0.1.0-rc.6 verified.
A. DSH plugin marketplace (recommended)
- Install dsh-plugin-marketplace if you don't have it yet (repo & one-line installer).
- Open Settings → DSH插件市场, search
dsh-agent-pricing, click 安装 / Install. - Restart DSH (relaunch
dsh web) and refresh the page.
B. dsh CLI (pnpm handles the GitHub URL)
dsh plugin --profile web add github:Way2LOose4/dsh-agent-pricing
# then restart the Web profile.
C. Manual
Clone into %DSH_HOME%/profiles/web/node_modules/dsh-agent-pricing and append to %DSH_HOME%/profiles/web/cordis.patch.yml:
- insert:
- id: agent-pricing
name: dsh-agent-pricing
Restart DSH. The Web profile's HMR is disabled — code changes only take effect on restart.
Usage
- Open any conversation — a compact cost line appears under the composer.
- Click CNY/USD to switch currency, click ▸ 图表 to open the today overlay.
- Ask your agent "how much have we spent?" — it can answer itself with
session_cost, and quote before-the-fact numbers withprice_estimate.
Configuration (optional)
Drop an agent-pricing.json into your workspace root to fully override the price table — same shape as prices.json in this repo:
{ "models": [{ "keys": ["my-model"], "label": "My Model", "input": 1.0, "output": 2.0 }] }
Optional per-row fields: cacheRead, cacheWrite, and tiers ({ "effectiveAt", "peakHoursUtc", "peak": {...}, "offPeak": {...} }). The file is re-read on activation and on every periodic refresh. Models that match no row cost 0 and are listed in the result's note.
HTTP endpoints & tools
| Endpoint | Method | Description |
|---|---|---|
/api/pricing/session-cost?sessionId=…¤cy=CNY |
GET | Full-session usage + cost |
/api/pricing/today?currency=CNY |
GET | Today's tokens/cost + hourly series + price source |
/api/pricing/estimate?model=…&input_tokens=…&output_tokens=… |
GET | Single-call estimate |
/api/pricing/debug |
GET | Price-source layers, sync state, tier rows |
/api/pricing/reload |
ANY | Force a price-table refresh |
Model tools registered: price_estimate, session_cost.
Keeping prices current
DeepSeek rows re-align automatically from the official page. For other vendors, update prices.json: every installation refetches it every 30 minutes. A workspace-level agent-pricing.json still wins when present.
Repo layout
├─ package.json bundle manifest (dsh.client + dsh.bundle.patch)
├─ cordis.patch.yml profile mount row
├─ prices.json repo-hosted price table (auto-fetched by installs)
├─ lib/
│ ├─ index.js host: routes, tools, price layers, official-page parser
│ └─ client.js browser: cost readout + today overlay
├─ test.mjs official-page parser + pricing math (peak/off-peak, CNY/USD)
└─ test-refresh.mjs price-layer precedence: file > official > repo > builtin
Run the tests against a saved copy of the official pricing page:
node test.mjs official-pricing.html
node test-refresh.mjs official-pricing.html
Limitations & security notes
- All figures are estimates: they mirror the official public price table, not your account's negotiated rates or remaining balance. DeepSeek exposes a real
/user/balanceendpoint; this plugin deliberately stays read-only on public data and needs no API key. - Cost history is computed from your local session logs; nothing leaves your machine except the price-table requests to the official docs page and this repo's raw
prices.json. - Only install plugins you trust — a plugin package is registered into the Web profile and loads with every DSH start.
Compatibility
| DSH version | Status |
|---|---|
| 0.1.0-rc.6 | ✅ verified |
License
MIT — see LICENSE.
安装
装一次目录插件,之后本站所有插件都能让 DeepSeek Harness 自动找、自动装:
dsh plugin add dshbase-catalog 然后对 agent 说「帮我装 dsh-agent-pricing」,它会在目录里找到并自动安装。文档:dshbase-catalog · 已验证场景包。
该插件是 GitHub 源码(未发 npm)——直接从仓库装:
Web profile:
dsh plugin --profile web add github:Way2LOose4/dsh-agent-pricing Headless(CLI)profile:
dsh plugin --profile headless add github:Way2LOose4/dsh-agent-pricing 实测报告
验证通过:从 GitHub 源码完成 L1 安装 + L2 加载 + L3 运行(dsh 0.1.0-rc.6)。
使用场景
让 agent 处理数据——解析、转换或分析——用真实数据集工作,而不只是文字。
适合谁
任务涉及表格、文件或数字、需要处理或分析的人。
二次开发建议
数据源和转换工具是缝——加格式、聚合或可视化输出。