@linxin666/dsh-usage 0.3.22 → 0.3.24

package/README.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   node scripts/verify-docs.mjs --write <dir>
-README.md: 2fe59d1c5b7e56aa1ec4a771334f6733d0e513c5
-README.zh.md: d0da2bc20e50639ac0d6133757b3bef7fd6536ec
+README.md: 1ec6e9c8d8f8f95a45777162471f3e1ee9a2768a
+README.zh.md: 6623cf55cbb4f3a72d90352af8ba52581057ee85

package/README.md

@@ -2,19 +2,19 @@
 
 English | [中文](README.zh.md)
 
-Usage statistics plugin for the dsh web GUI: per-provider balance and coding-plan quota detection plus a live token usage ledger, with a dedicated pet announcement bubble for the current provider.
+Usage statistics plugin for the dsh web GUI: per-provider balance and coding-plan quota detection plus a live token usage ledger, with the current session provider's today usage on the sidebar entry.
 
 ## What it does
 
 The plugin runs a host-side service and a first-level settings section (使用统计, directly below the Workshop entry):
 
-- **Usage tab (用量)**: today's token totals per bucket (input / output / cache read / cache write, disjoint as the provider reports them) with per-provider and per-model breakdown, the last 30 days as a horizontal provider/model bar chart, and the balance of every configured provider. For the official DeepSeek family the tab also shows the current peak/off-peak pricing period (Beijing weekdays 09:00-12:00 and 14:00-18:00 are peak, billed at double) and today's estimated spend in CNY. The ledger folds live `session/event` streams (`request/header` route attribution plus `assistant/message` usage) into `$DSH_HOME/dsh-usage/usage-ledger.json` with day-based retention; counting starts when the plugin is first enabled.
-- **Plans tab (个人套餐)**: coding-plan quota windows for every configured provider that exposes one — used percent and reset time per window (Kimi For Coding 5h/week, GLM Coding Plan 5h/week, OpenCode Go rolling/weekly/monthly, MiniMax 5h/week, Codex / ChatGPT subscription 5h/week). Providers without a real plan or subscription (DeepSeek, ZenMux, Moonshot, OpenRouter, SiliconFlow) never appear on this tab; their balance shows on the usage tab instead.
-- **Token Bank tab (Token 银行)**: the whale-yuan voucher (鲸元券) minted from the official DeepSeek family's ledger at an anti-inflation exchange rate of 1,000 tokens per whale yuan. The tab sums the family's retained-ledger tokens (the `deepseek` catalog alias and the live `deepseek-official` route combined), stamps the minted face value onto the banknote artwork with a serial line carrying the minting window, and offers a save-image button plus a native share button when the browser supports sharing files. The spend line shows the real CNY total observed from the official balance once available — balance decreases accrue as spend, top-ups never count — and falls back to the fold-time estimate before the first observation. The minted total prefers the host's whole-ledger aggregate and falls back to the last 30 days on an older host; with no official DeepSeek usage the tab shows its empty state. The on-note text is locale-neutral (digits, latin captions, ISO dates) and the export is rendered locally in the browser.
-- **Pet linkage**: the pet renders a dedicated announcement bubble (own glass style, tone accent, mini quota meter) following the current session provider. Plan providers (Kimi, GLM, Codex subscription, ...) show their tightest percent window; the official DeepSeek family shows today's estimated spend with the current peak/off-peak period and the account balance. When the session provider has no announceable probe fact (a relay station or local runtime without an adapter, failing probes, percent-less windows), the bubble falls back to that provider's live usage today (tokens and call count); with neither fact nor usage it stays silent. `bubbleMode` controls it: always (refreshed every poll with a poll-cadence TTL, so the bubble stays visible) / on change / off.
+- **Usage tab (用量)**: today's token totals per bucket (input / output / cache read / cache write, disjoint as the provider reports them) with per-provider and per-model breakdown, the last 30 days as a horizontal provider/model bar chart, and the balance of every configured provider (one row per adapter family: a dormant catalog alias shadowed by the family's live route no longer renders its own balance row). For the official DeepSeek family the tab also shows the current peak/off-peak pricing period (Beijing weekdays 09:00-12:00 and 14:00-18:00 are peak, billed at double) and today's estimated spend in CNY. The ledger folds live `session/event` streams (`request/header` route attribution plus `assistant/message` usage) into `$DSH_HOME/dsh-usage/usage-ledger.json` with day-based retention; counting starts when the plugin is first enabled.
+- **Plans tab (个人套餐)**: coding-plan quota windows for every configured provider that exposes one — used percent and reset time per window (Kimi For Coding 5h/week, GLM Coding Plan 5h/week/month, OpenCode Go rolling/weekly/monthly, MiniMax 5h/week, Codex / ChatGPT subscription 5h/week). Providers without a real plan or subscription (DeepSeek, ZenMux, Moonshot, OpenRouter, SiliconFlow) never appear on this tab; their balance shows on the usage tab instead.
+- **Token Bank tab (Token 银行)**: the whale-yuan voucher (鲸元券) minted from the official DeepSeek family's ledger at an anti-inflation exchange rate of 1,000,000 tokens per whale yuan. The tab sums the family's retained-ledger tokens (the `deepseek` catalog alias and the live `deepseek-official` route combined), stamps the minted face value onto the banknote artwork with a serial line carrying the minting window, and offers a save-image button plus a native share button when the browser supports sharing files. The spend line shows the real CNY total observed from the official balance once available — balance decreases accrue as spend, top-ups never count — and falls back to the fold-time estimate before the first observation. The minted total prefers the host's whole-ledger aggregate and falls back to the last 30 days on an older host; with no official DeepSeek usage the tab shows its empty state. The on-note text is locale-neutral (digits, latin captions, ISO dates) and the export is rendered locally in the browser.
+- **Sidebar panel**: a usage entry in the sidebar (directly below the family plugin entries) expands into a collapsible panel showing the same overview document — plan providers with their quota windows and tightest percent, balance-only providers with their remaining amount. The entry row carries the panel's controls (a refresh button with a short cooldown and a collapse chevron). Collapsed, a one-line strip under the row shows the current session provider's today usage (tokens, calls, and the priced cost when the family is priced); with no session provider or no usage today the strip stays silent. The collapsed state is remembered in `localStorage` (`dsh-usage.sidebar.collapsed`). Polling runs at 10 s while expanded and 30 s while collapsed, and pauses whenever the page is hidden. The panel reads `/api/dsh-usage/overview` through the same family trust fence as the settings section: loopback always passes, and a browser on a paired LAN device passes when `dsh-remote-web-ui` is loaded — an unpaired LAN request still gets 403.
 - Probes run entirely host-side on a poll cycle (default 60 s, manual refresh button); API keys are resolved through the harness credential seam (`llm-pi-ai` records, `apiKeyEnv` references) and never reach the browser.
 
-Supported balance endpoints: DeepSeek (the official live route `deepseek-official` and the catalog alias `deepseek` both resolve), Moonshot (CN/international), OpenRouter, SiliconFlow (CN/international), ZenMux. Supported plan endpoints: Kimi For Coding, GLM Coding Plan (CN via open.bigmodel.cn, international via api.z.ai), OpenCode Go, MiniMax, Codex / ChatGPT subscription (OAuth access token from the pi-ai grant; a stale token shows an error until the harness next refreshes it). Providers without a programmatic endpoint (Qwen token plans, OpenCode Zen PAYG, Anthropic, OpenAI) are listed without facts.
+Supported balance endpoints: DeepSeek (the official live route `deepseek-official` and the catalog alias `deepseek` both resolve), Moonshot (CN/international), OpenRouter, SiliconFlow (CN/international), ZenMux. Supported plan endpoints: Kimi For Coding, GLM Coding Plan (CN via open.bigmodel.cn, international via api.z.ai — the `zai` route the pi-ai catalog registers, plus the `zai-coding` / `zai-coding-cn` aliases; 5-hour, weekly and monthly windows, token-metered and credit-metered plans), OpenCode Go, MiniMax, Codex / ChatGPT subscription (OAuth access token from the pi-ai grant; a stale token shows an error until the harness next refreshes it). Providers without a programmatic endpoint (Qwen token plans, OpenCode Zen PAYG, Anthropic, OpenAI) are listed without facts.
 
 ### Spend estimate scope
 
@@ -48,7 +48,6 @@ Restart `dsh web` for the host half; the client half applies on refresh. The sec
 | --- | --- | --- |
 | `enabled` | `true` | Master switch; off = no listeners, no probes, no routes |
 | `pollIntervalSec` | `60` | Provider probe cycle (30-3600 s, hot-swappable) |
-| `bubbleMode` | `always` | Pet announcement bubble: `always` (fresh every poll) / `change` (only when the value changes) / `off` |
 | `retainDays` | `180` | Ledger retention in local days (7-730) |
 
 ## Known limitations
@@ -58,4 +57,3 @@ Restart `dsh web` for the host half; the client half applies on refresh. The sec
 - A failed probe keeps the previous fact visible and reports the error line; providers may rate-limit aggressive polling.
 - The voucher's face value covers the ledger's retention window (`retainDays`): pruned days drop out of both the trend chart and the note.
 - The real-spend watch starts at the first official balance observation: earlier consumption is not backfilled, and only observed balance decreases accrue (a balance rise is a top-up and never counts).
-- The pet bubble requires the dsh-pet plugin; without it the section works unchanged.

package/README.zh.md

@@ -2,19 +2,19 @@
 
 [English](README.md) | 中文
 
-dsh Web GUI 的使用统计插件：多 provider 余额与编程套餐用量检测，外加实时 token 用量台账，并通过宠物插件的专用公告气泡展示当前提供方状态。
+dsh Web GUI 的使用统计插件：多 provider 余额与编程套餐用量检测，外加实时 token 用量台账，并在侧栏入口展示当前会话提供方的今日用量。
 
 ## 功能
 
 插件由宿主侧服务与设置页一级分区（使用统计，位于创意工坊下方）组成：
 
-- **用量页签**：今日 token 分桶合计（输入 / 输出 / 缓存读 / 缓存写，按 provider 上报口径互不相加），分 provider 与模型细分，近 30 天以「提供方-模型」水平条形图展示，以及所有已配置 provider 的余额。对 DeepSeek 官方路由，页签还展示当前峰谷计价时段（北京时间工作日 09:00-12:00、14:00-18:00 为高峰，按双倍计费）与今日消费估算（CNY）。台账从 `session/event` 实时流折叠（`request/header` 归因路由 + `assistant/message` 用量），持久化到 `$DSH_HOME/dsh-usage/usage-ledger.json`，按本地日保留；统计自插件首次启用起计。
-- **个人套餐页签**：每个已配置且暴露套餐端点的 provider 的配额窗口——已用百分比与重置时间（Kimi For Coding 5 小时/每周、GLM 编程计划 5 小时/每周、OpenCode Go 滚动/每周/每月、MiniMax 5 小时/每周、Codex / ChatGPT 订阅 5 小时/每周）。没有真实套餐/订阅体系的厂商（DeepSeek、ZenMux、Moonshot、OpenRouter、SiliconFlow）不出现在此页签，其余额显示在用量页签。
-- **Token 银行页签**：以 DeepSeek 官方家族的台账用量铸造「鲸元券」，汇率按 1000 tokens 兑 1 鲸元（防膨胀）。页签把该家族保留台账内的 token 总量（`deepseek` 目录别名与运行时路由 `deepseek-official` 合并计算）折算成票面面额印在钞票图上，附带走票窗口的序列号行，并提供保存图片按钮与（浏览器支持文件分享时的）系统分享按钮。消费行优先展示从官方余额观测到的真实累计花费——余额下降即计为消费，充值上涨不计——首次观测到下降之前回落为折叠时刻估算。铸造总量优先取宿主的全台账聚合，旧宿主回落为近 30 天；无官方用量时展示空状态。票面文字与语言无关（数字、拉丁小字、ISO 日期），导出图在浏览器本地渲染。
-- **宠物联动**：宠物渲染一只专用公告气泡（独立玻璃样式、色调描边、微型配额计量条），跟随当前会话提供方。套餐类 provider（Kimi、GLM、Codex 订阅等）展示最紧的百分比窗口；DeepSeek 官方路由展示今日消费估算、当前峰谷时段与账户余额。会话提供方没有可公告的探测事实时（无适配器的中转站或本地运行时、探测失败、无百分比窗口），气泡回落为该提供方今日的实时用量（tokens 与调用次数）；事实与用量皆无时保持沉默。`bubbleMode` 控制行为：常驻（每次轮询即刷新，TTL 随轮询周期走，气泡保持可见）/ 仅变化时 / 关闭。
+- **用量页签**：今日 token 分桶合计（输入 / 输出 / 缓存读 / 缓存写，按 provider 上报口径互不相加），分 provider 与模型细分，近 30 天以「提供方-模型」水平条形图展示，以及所有已配置 provider 的余额（同一适配器家族只渲染一行：被该家族运行时路由遮蔽的休眠目录别名不再单独出余额行）。对 DeepSeek 官方路由，页签还展示当前峰谷计价时段（北京时间工作日 09:00-12:00、14:00-18:00 为高峰，按双倍计费）与今日消费估算（CNY）。台账从 `session/event` 实时流折叠（`request/header` 归因路由 + `assistant/message` 用量），持久化到 `$DSH_HOME/dsh-usage/usage-ledger.json`，按本地日保留；统计自插件首次启用起计。
+- **个人套餐页签**：每个已配置且暴露套餐端点的 provider 的配额窗口——已用百分比与重置时间（Kimi For Coding 5 小时/每周、GLM 编程计划 5 小时/每周/每月、OpenCode Go 滚动/每周/每月、MiniMax 5 小时/每周、Codex / ChatGPT 订阅 5 小时/每周）。没有真实套餐/订阅体系的厂商（DeepSeek、ZenMux、Moonshot、OpenRouter、SiliconFlow）不出现在此页签，其余额显示在用量页签。
+- **Token 银行页签**：以 DeepSeek 官方家族的台账用量铸造「鲸元券」，汇率按 100 万 tokens 兑 1 鲸元（防膨胀）。页签把该家族保留台账内的 token 总量（`deepseek` 目录别名与运行时路由 `deepseek-official` 合并计算）折算成票面面额印在钞票图上，附带走票窗口的序列号行，并提供保存图片按钮与（浏览器支持文件分享时的）系统分享按钮。消费行优先展示从官方余额观测到的真实累计花费——余额下降即计为消费，充值上涨不计——首次观测到下降之前回落为折叠时刻估算。铸造总量优先取宿主的全台账聚合，旧宿主回落为近 30 天；无官方用量时展示空状态。票面文字与语言无关（数字、拉丁小字、ISO 日期），导出图在浏览器本地渲染。
+- **侧栏面板**：侧栏在家族插件入口下方提供「用量」入口，展开为可折叠面板，读取与设置分区相同的 overview 文档——套餐类 provider 展示配额窗口与最紧百分比，仅余额类 provider 展示剩余额度。入口行本身承载面板控件（带短冷却的刷新按钮与折叠箭头）。折叠态下，入口行下方的一行摘要条展示当前会话提供方的今日用量（tokens、调用次数，计价家族附带费用）；无会话提供方或今日无用量时摘要条保持沉默。折叠状态记在 `localStorage`（`dsh-usage.sidebar.collapsed`）。轮询节奏为展开 10 秒、折叠 30 秒，页面隐藏时一律暂停。面板经与设置分区相同的家族信任围栏读取 `/api/dsh-usage/overview`：回环地址始终放行，已配对的局域网设备在装有 `dsh-remote-web-ui` 时放行，未配对的局域网请求仍返回 403。
 - 探测完全在宿主侧按轮询周期执行（默认 60 秒，可手动刷新）；API key 经宿主凭据缝解析（`llm-pi-ai` 记录、`apiKeyEnv` 引用），永不进入浏览器。
 
-支持的余额端点：DeepSeek（官方运行时路由 `deepseek-official` 与目录别名 `deepseek` 均可解析）、Moonshot（国内/国际）、OpenRouter、SiliconFlow（国内/国际）、ZenMux。支持的套餐端点：Kimi For Coding、GLM 编程计划（国内 open.bigmodel.cn、国际 api.z.ai）、OpenCode Go、MiniMax、Codex / ChatGPT 订阅（OAuth access token 取自 pi-ai grant；token 过期时显示错误行，宿主下次跑 Codex 请求自动刷新后恢复）。没有程序化端点的 provider（Qwen token 套餐、OpenCode Zen 按量、Anthropic、OpenAI）仅列出，不展示数据。
+支持的余额端点：DeepSeek（官方运行时路由 `deepseek-official` 与目录别名 `deepseek` 均可解析）、Moonshot（国内/国际）、OpenRouter、SiliconFlow（国内/国际）、ZenMux。支持的套餐端点：Kimi For Coding、GLM 编程计划（国内 open.bigmodel.cn、国际 api.z.ai——即 pi-ai 目录注册的 `zai` 路由，另兼容 `zai-coding` / `zai-coding-cn` 别名；覆盖 5 小时 / 每周 / 每月窗口与 token 制、积分制套餐）、OpenCode Go、MiniMax、Codex / ChatGPT 订阅（OAuth access token 取自 pi-ai grant；token 过期时显示错误行，宿主下次跑 Codex 请求自动刷新后恢复）。没有程序化端点的 provider（Qwen token 套餐、OpenCode Zen 按量、Anthropic、OpenAI）仅列出，不展示数据。
 
 ### 消费估算口径
 
@@ -48,7 +48,6 @@ pnpm add @linxin666/dsh-usage
 | --- | --- | --- |
 | `enabled` | `true` | 总开关；关闭后不监听、不探测、不注册路由 |
 | `pollIntervalSec` | `60` | provider 探测周期（30-3600 秒，热切换） |
-| `bubbleMode` | `always` | 宠物公告气泡：`always`（每次轮询即刷新）/ `change`（仅数值变化时）/ `off` |
 | `retainDays` | `180` | 台账按本地日保留天数（7-730） |
 
 ## 已知限制
@@ -58,4 +57,3 @@ pnpm add @linxin666/dsh-usage
 - 探测失败时保留上一次数据并展示错误行；过于频繁的轮询可能触发 provider 限流。
 - 鲸元券面额只覆盖台账保留窗口（`retainDays`）内的用量：被清理的天同时退出趋势图与票面。
 - 真实花费观测从第一次官方余额观测开始：更早的消费不回补，且只有观测到的余额下降才累计（余额上涨是充值，永不计费）。
-- 宠物气泡需要 dsh-pet 插件；未安装时分区功能不受影响。