LLM Quota Checker

A desktop GUI application that monitors the availability of LLM API accounts across multiple providers. It checks quota limits automatically on a configurable interval, shows live countdowns for rate-limited models, and helps you discover which models are available and free on your API keys.

---

Features

Account Monitoring

• Monitors multiple API accounts and models simultaneously
• Configurable check interval (default: 60 minutes)
• Live countdown timer when a model is rate-limited — skips re-checking until the cooldown expires
• "Check Now" button to force an immediate check regardless of cooldowns
• Persistent storage: all accounts and their last-known status survive restarts

Status Classification

Status	Meaning
✔ Available	Quota is currently usable
✔ Available ★	Available and confirmed free ($0 per token)
✖ Limited	HTTP 429 – daily/hourly quota reached, cooldown active
💳 No Balance	HTTP 402 – model requires credits or a paid account
⚠ Error	Other error (connection, 403, etc.)

Model Scanner

• Loads the full model list from the OpenRouter API (no auth required, ~300+ models)
• For NVIDIA NIM: fetches the model list directly from /v1/models using your API key
• Tests each model in parallel (4 concurrent workers) against your API key
• Shows live results as they come in, with price data from OpenRouter
• Distinguishes between confirmed-free models (:free suffix), assumed-free ($0 price), and models that return 402 despite a $0 price listing
• Filter results by status; select individual models or bulk-add all available ones as accounts
• Custom model IDs can be added to the test list manually

System Tray

• Minimizes to the system tray (both via the X button and the minimize button)
• Tray icon color reflects overall status: 🟢 all available · 🟡 some limited · 🔴 all limited
• Tooltip shows per-account status with remaining cooldown time
• Right-click menu: Open, Check Now, Quit

Account Management

• Each account is identified by an internal UUID — the display name (provider) can appear multiple times without conflicts
• Add, edit, delete accounts via dialog; API key field is masked with a visibility toggle
• Test connection directly in the add/edit dialog before saving
• Sortable columns in both the main table and the scan results table

---

Supported Providers

Provider	Completions URL	Model Discovery
Cline	https://api.cline.bot/api/v1/chat/completions	via OpenRouter API
OpenRouter	https://openrouter.ai/api/v1/chat/completions	via OpenRouter API
NVIDIA NIM	https://integrate.api.nvidia.com/v1/chat/completions	via /v1/models (auth required)
Any OpenAI-compatible API	custom URL	manual model IDs

---

Installation

Requirements: Python 3.10+

pip install requests pystray pillow

Or use the included Windows batch file which installs dependencies automatically.

Run:

python llm_quota_checker_gui.py

Windows: double-click LLM_Quota_Checker_Starten.bat

To start automatically on Windows login, place a shortcut to the .bat file in:

shell:startup

---

Usage

Adding accounts manually

1. Click ＋ Add in the toolbar
2. Fill in:
   • Name — the provider label (e.g. Cline, OpenRouter). Not unique — multiple accounts can share the same name.
   • API Key — your secret key for this provider
   • Model — the model ID in provider/model-name format (e.g. xiaomi/mimo-v2.5)
   • URL — the completions endpoint
3. Click Test Connection to verify before saving

Discovering available models (Scan)

1. Select an existing account in the table (optional — pre-fills key and URL)
2. Click 🔍 Scan Models in the toolbar
3. Enter API Key and URL if not pre-filled
4. Click ▶ Start Scan
   • The full model list is fetched from OpenRouter (or NVIDIA NIM)
   • Each model is tested against your API key in parallel
   • Results appear live in the table with status and price info
5. Select individual models (Ctrl+click / Shift+click) or leave nothing selected to add all available ones
6. Click ＋ Add Selection as Account

Cooldown logic

When a model returns HTTP 429, the response is parsed for a wait time (e.g. Try again in 3h 22m). The model is then skipped automatically until that time has passed. The main table shows a live countdown: Cooldown – noch 2h 47m (free at 15:22:00).

Check Now (⟳) always bypasses cooldowns and forces a full re-check of all accounts.

HTTP 402 responses (no balance / credits exhausted) set a 24-hour skip — retrying every hour serves no purpose since balance doesn't replenish automatically.

---

Configuration

Settings are stored in ~/.llm_quota_checker.json and are loaded automatically on startup.

{
  "interval_minutes": 60,
  "accounts": [
    {
      "id": "a1b2c3d4",
      "name": "Cline",
      "api_key": "sk_...",
      "model": "xiaomi/mimo-v2.5",
      "url": "https://api.cline.bot/api/v1/chat/completions",
      "status": "ok",
      "detail": "",
      "last_check": "14:22:01",
      "retry_until": null
    }
  ]
}

The interval can also be changed at runtime in the toolbar — press Enter to apply.

---

Free Model Detection

Free model detection is fully dynamic — no hardcoded model lists.

OpenRouter / Cline: A model is considered free if:

1. Its ID ends with :free (explicit OpenRouter free tier), or
2. Its price from the OpenRouter API is $0 per token

If the actual API call returns HTTP 402, the model is reclassified as 💳 Kostenpflichtig regardless of the listed price. This catches models that are listed as $0 but still require a paid account on the target provider.

NVIDIA NIM: NVIDIA uses a credits system. New accounts receive 1,000 free inference credits. After that, all models require credits. The app shows Credits required for NVIDIA models and handles their different error format ({"status": 429, "title": "Too Many Requests"}).

---

Dependencies

Package	Purpose
requests	HTTP calls to LLM APIs and OpenRouter
pystray	System tray icon
Pillow	Tray icon image generation
tkinter	GUI (included with Python)

---

Notes

• The app sends a minimal test request (max_tokens: 1, message: "Hi") to check quota — this uses negligible tokens
• Parallel scan workers are capped at 4 to avoid triggering rate limits during discovery
• The tray tooltip is limited to 127 characters (Windows system limit)