DeepSeek Harness Plugin

ljr282341583/dsh-convo-cost

Stars ★ 0 Category Usage & Billing Added 2026-08-18 npm dsh-convo-cost

Real-time cost tracking for the current DSH conversation: a composer-dock line showing the running token usage and DeepSeek official price estimate (base and peak/valley pricing).

Install

# from npm (prebuilt)

dsh plugin --profile web add dsh-convo-cost

# from GitHub (first run asks for allowBuilds approval — follow the hint, retry)

dsh plugin --profile web add github:ljr282341583/dsh-convo-cost

Any plugin you install runs third-party code with your own permissions — it can read your files, use your credentials, and reach the network, and tool approvals don’t sandbox it. GitHub-sourced plugins also run build scripts at install time. Only install sources you trust, and pin a commit (github:owner/repo#sha).

README

Real-time token usage and cost tracking for the current DSH conversation: a status line under the composer shows ¥x.xx · xx tok for this conversation, and hovering reveals call count / input (cache-hit included) / output (reasoning included) / cost / pricing regime / model details.

  • Pricing: DeepSeek official price list (CNY per 1M tokens), cross-checked against api-docs.deepseek.com/zh-cn/quick_start/pricing
  • Automatic regime switch: base prices before 2026-08-17 00:00 (Beijing time), peak/valley prices after
  • Peak hours (Beijing time 09:00-12:00 and 14:00-18:00) are billed at the peak rate automatically
  • The model name is read from request headers; unknown models are billed as ¥0

Install

dsh plugin --profile web add dsh-convo-cost

Restart dsh web, and the conversation status line shows 本对话 ¥x.xx · xx tok.

Effect

Status line Hover details
本对话 ¥1.47 · 20.9M tok 152 calls · input 20,812,672 tok (cache hit 20,812,672) · output 118,945 tok (reasoning …) · ≈ ¥1.47 · pricing: base · model: deepseek-v4-flash

How it works

  • Host side (lib/index.js): subscribes to the session event stream through the sessionProjections service, accumulates tokens on each assistant/message event's exact usage, prices them against the official table, and registers the convoCost projection.
  • Client side (lib/client.js): injects into the conversation.composer.dock slot and renders the cost line with useProjection("convoCost"), refreshing live with the conversation.

Pricing table (deepseek-v4-flash, CNY per 1M tokens)

Item Base (before 08-17) Peak/valley · off-peak Peak/valley · peak
Input · cache hit 0.02 0.05 0.10
Input · cache miss 1.0 1.5 3.0
Output 2.0 4.5 9.0

deepseek-v4-pro and the full peak/valley table live in the PRICING table in lib/index.js; prices follow the official DeepSeek page — update the table there when the official prices change.

Test

npm test

The projection's token accounting and billing are verified with a simulated event stream (no network, no real calls).

FAQ

Symptom Cause / fix
Status line missing Restart dsh web after install; confirm the plugin joined dsh.profile.bundles (dsh plugin add reconciles it automatically)
Always shows ¥0.00 The model is not in the official price table (unknown models bill at ¥0); wait for the first assistant/message event carrying usage
Amount differs from the bill Estimate only: priced from the request-header model name and the official table, excluding discounts/promotions/platform markup; the official bill is authoritative
Prices changed officially Update the PRICING table and EFFECTIVE_AT in lib/index.js, then republish the npm package and update
Hover details empty The projection needs at least one assistant/message event with usage; new sessions start from 0

License

MIT

Content from the project README on GitHub ↗

Links

GitHub repository ↗ npm package ↗ Install via dsh-market ↗

More in this category

View the whole category →