mmx-quota-tool MiniMax token-plan quota dock for DeepSeek Harness (DSH) web. Shows a drop-shape icon + 5h usage % in the conversation input area (next to the model selector). Click for a detail panel listing every model's 5h & weekly usage, reset countdown, and absolute numbers. Auto-hides when the active default model is not a MiniMax model. ## Install sh dsh plugin --profile web add github:mtty-ai/mmx-quota-tool Then restart dsh web once. ### Manual install (development) sh git clone https://github.com/mtty-ai/mmx-quota-tool cd mmx-quota-tool # copy or link the package into your profile's node_modules cp -R . ~/.dsh/profiles/web/node_modules/mmx-quota-tool/ Then add the row to ~/.dsh/profiles/web/cordis.patch.yml: yaml - insert: - id: mmx-quota-tool name: mmx-quota-tool ## Configuration This plugin requires a MiniMax API key. Export it before launching dsh web: sh export MMX_API_KEY='sk-cp-your-key-here' dsh web Or put it in your shell rc (~/.zshrc, ~/.bashrc). Optional environment variables: | Var | Default | Purpose | | --------------------- | ------------------------------------ | -------------------------------------- | | MMX_API_KEY | (required) | MiniMax Bearer + x-api-key | | MMX_REGION | cn | cn (api.minimaxi.com) or global (api.minimax.io) | | MMX_REFRESH_MS | 60000 | Host poll interval against upstream | | MMX_POLL_INTERVAL_MS| 5000 | (client bundle) Client fetch interval| The client bundle polls the host every 5 s so model-switch visibility is within 5 s; the actual upstream fetch runs every MMX_REFRESH_MS. ## Behavior - The dock appears in the conversation input area (conversation.input.left), adjacent to the model selector. - Icon is a drop/water shape, filled by 5h usage % (lower bar = better): - < 50% green - < 80% yellow - ≥ 80% red - Click toggles a detail panel; mouse-leave the dock or panel closes it. - Shift+click forces an immediate upstream refresh. - The dock is hidden entirely when the active default model is not a MiniMax model — matched providers minimax / minimax-cn / minimax-global / MiniMax* or model prefixes MiniMax- / minimax-. - Model name labels follow the active DSH UI locale: - zh: 通用 / 视频 / 图像 / 音频 / 语音 / 音乐 / 视觉 / 向量 / 实时 / 长文本 - en: General / Video / Image / Audio / Speech / Music / Vision / Embedding / Realtime / Long context - The dock label, panel header / headline / footer hint all switch language when the user changes DSH's interface language. - The default is zh (Chinese) — the language the upstream token-plan model names resolve to most naturally. If the locale service is unavailable the dock still renders. ## Architecture | Half | File | Responsibility | | ------ | --------------------- | ---------------------------------------------------- | | Host | src/index.js | 60 s upstream poll + exposes two HTTP routes | | Client | client/client.js | 5 s poll + mounts the dock UI in conversation.input.left | The host registers two routes on the profile's webServer: - GET /api/mmx-quota-tool/quota — current snapshot - POST /api/mmx-quota-tool/refresh — force refresh Returns { rows, lastFetchedAt, lastError, lastStatus, model, isMmx, refreshIntervalMs }. The client polls quota and renders null when isMmx === false, so switching to a non-MiniMax model hides the dock within 5 s. ## Files . ├── README.md ├── LICENSE ├── package.json # dsh.bundle + dsh.client declaration + exports ├── cordis.patch.yml # dsh bundle patch (registers the row) ├── src/index.js # host: HTTP routes + 60s upstream poll └── client/client.js # browser: 5s poll + React UI + injected <style> ## License MIT