Coding Plan Dashboard

A single-page dashboard that aggregates the coding-plan quotas of multiple AI providers (Codex, MiniMax, Volcengine AgentPlan / CodingPlan, Kimi Code, LongCat, Qianwen AI and Google AI Gemini 3.5 Flash) into one self-hosted UI. Each provider supports multiple accounts with independent labels.

一个部署在局域网中的 Coding Plan 配额看板，支持 Codex、MiniMax、火山方舟 CodingPlan / AgentPlan、Kimi Code、LongCat、千问 AI 和 Google AI (Gemini 3.5 Flash)。支持同一平台 多账号，每个账号可添加备注。

Features / 功能

• Auto-detects the data source from the imported raw curl URL — no manual selection needed. / 根据原始 curl URL 自动识别数据来源，不需要手工选择产品。
• The server re-executes saved requests on page open, manual refresh, or auto refresh; results are persisted as JSON on the host so they survive container restarts. / 页面打开、刷新或点击刷新按钮时由服务器重新执行已保存的请求；配额结果保存到服务器端 JSON 文件，容器重启后自动恢复。
• First paint shows the cached snapshot then silently refreshes to the latest data; successful refreshes do not pop up any dialog.
• Multi-account support: each imported curl is stored as a separate account with its own ID, label, edit, delete (with double confirm), and reorder controls.
• Per-account Volcengine AK/SK configuration (no curl required); uses HMAC-SHA256 V4 signing against GetCodingPlanUsage / GetAgentPlanAFPUsage.
• Codex supports both the official /usage + /rate-limit-reset-credits endpoints and a NewAPI replacement (/api/channel/{channelId}/codex/usage[/reset-credits]). The server merges them into a single Codex card and skips the official requests when a NewAPI variant is configured.
• All progress bars show one-decimal usage percentage; each card shows a live countdown to the next reset in the top-right corner.
• Self-contained single-file browser UI (index.html) with a custom SVG logo and no external favicon dependency.

File Layout

index.html          Single-file frontend
server.py           Static file serving + JSON persistence + restricted curl execution
parser.js           Parser implementation used by Node tests
docker-compose.yml  Container deployment configuration
tests/              Parser and server unit tests
AGENTS.md           Coding-agent instructions for this repository

Quick Start (Docker)

Requirements

• Linux host / NAS with Docker Engine and Docker Compose v2.
• The host must be able to reach Codex, MiniMax and Volcengine APIs.
• If you import a Codex curl with --proxy, the proxy must be reachable from the Docker host.
• Default listen port is 8080; change PORT in docker-compose.yml if occupied.

1. Prepare the host directory

sudo mkdir -p /docker/coding_plan_quota_dashboard/data
sudo chmod 700 /docker/coding_plan_quota_dashboard/data
sudo cp index.html server.py docker-compose.yml /docker/coding_plan_quota_dashboard/

2. Start the stack

cd /docker/coding_plan_quota_dashboard
sudo docker compose up -d
sudo docker compose ps
sudo docker logs --tail=100 coding-plan-quota-dashboard
curl -fsS -o /dev/null -w 'HTTP %{http_code}\n' http://127.0.0.1:8080/

3. Portainer Stack

In Portainer choose the target endpoint, Stacks → Add stack, name it coding-plan-quota-dashboard, paste the contents of docker-compose.yml and click Deploy the stack. Because Portainer uses host bind mounts, index.html, server.py and an empty data/ directory must already exist on the host.

Updating

sudo cp index.html /docker/coding_plan_quota_dashboard/index.html
sudo cp server.py /docker/coding_plan_quota_dashboard/server.py
sudo docker restart coding-plan-quota-dashboard

For Compose-level changes use Portainer's Update the stack, or:

cd /docker/coding_plan_quota_dashboard
sudo docker compose up -d --force-recreate

Persistent Data & Backups

/docker/coding_plan_quota_dashboard/data/requests.json    # imported curl, contains secrets
/docker/coding_plan_quota_dashboard/data/credentials.json # per-account Volcengine AK/SK
/docker/coding_plan_quota_dashboard/data/snapshot.json    # latest quota snapshot
/docker/coding_plan_quota_dashboard/data/results.json     # per-account cached API responses

Backup:

sudo tar -czf "coding_plan_quota_dashboard-data-$(date +%Y%m%d-%H%M%S).tar.gz" \
  -C /docker/coding_plan_quota_dashboard data

Restore:

cd /docker/coding_plan_quota_dashboard
sudo docker compose down
sudo tar -xzf coding_plan_quota_dashboard-data-YYYYMMDD-HHMMSS.tar.gz
sudo docker compose up -d

Treat requests.json and credentials.json as secrets: do not commit them, and never publish the backup file to a public download location.

Importing Requests

Paste the full raw curl in the page and click 保存 curl 并刷新. The server auto-classifies by URL:

• chatgpt.com/backend-api/wham/usage → Codex usage
• chatgpt.com/backend-api/wham/rate-limit-reset-credits → Codex reset credits
• <newapi-host>/api/channel/{channelId}/codex/usage → NewAPI Codex usage
• <newapi-host>/api/channel/{channelId}/codex/usage/reset-credits → NewAPI Codex reset credits
• www.minimaxi.com → MiniMax
• GetCodingPlanUsage → Volcengine CodingPlan
• GetAgentPlanAFPUsage → Volcengine AgentPlan (current endpoint)
• GetAgentPlanUsageDetails → Volcengine AgentPlan (legacy endpoint, still supported)
• www.kimi.com/.../GetSubscriptionStats → Kimi Code
• longcat.chat/api/pay/quota/metering/token-packs/summary → LongCat
• cs-data.qianwenai.com/.../data/api.json (contains tokenplan) → Qianwen AI TokenPlan
• fetchAvailableModels (Gemini 3.5 Flash) → Google AI (Antigravity)

Codex official and NewAPI endpoints can coexist; the server merges them into one card. curl -sS, --proxy and --insecure flags are translated into Python HTTPS requests — the server never spawns a shell to run curl.

If /usage returns secondary_window: null, the page keeps the 5-hour row visible and shows "接口未返回" instead of fabricating usage.

Security Notes

• The server does not shell out; imported curl is parsed and re-issued via Python urllib-style HTTPS requests.
• The host allowlist defaults to chatgpt.com, www.minimaxi.com, console.volcengine.com, www.kimi.com, longcat.chat, cs-data.qianwenai.com; additional NewAPI hosts can be enabled via the NEWAPI_HOSTS (comma-separated) environment variable and are accepted only for the Codex /usage and /usage/reset-credits paths.
• Google AI uses OAuth client pairs loaded from the GOOGLE_AI_CLIENTS environment variable as a JSON array of [client_id, client_secret] pairs. When unset, the server boots with three REDACTED_* placeholders and the Google AI refresh call will fail until you provide real values; this keeps credentials out of source code.
• NewAPI is reached over plain HTTP because it is a LAN service — do not extend this rule to arbitrary external HTTP endpoints.
• When a codexNewApi request is saved the refresh automatically skips the official codexUsage request, and likewise for codexNewApiCredits → codexCredits. This avoids repeated 401 noise when the official token is stale.
• Raw curl (Token / Cookie / Digest) is stored in requests.json in plaintext, and Volcengine AK/SK is stored in credentials.json in plaintext. Never commit real credentials; the README and tests use redacted samples only.
• Restrict 8080 to a trusted LAN and rotate credentials immediately on leak.

Tests

node --test tests/parser.test.js
PYTHONPATH=. python3 -m unittest discover -s tests -p 'test_*.py'
python3 -m py_compile server.py

CI runs the same commands on every push and pull request; see .github/workflows/test.yml.

License

MIT — see LICENSE for the full text.