LLM Limits Tracker

Local dashboard for LLM API rate limits — across Claude, GPT, Gemini, OpenRouter, Groq, Fireworks, Mistral and Together.

Stop opening eight different consoles. Run one command, see every quota in one screen.

Why this exists

If you build with multiple LLMs, you already know the routine: production hits a 429, you alt-tab to one console, then another, then a third, trying to figure out which provider is the bottleneck. The big observability platforms (Helicone, Langfuse, LiteLLM) solve this for teams — but they're heavy, require code instrumentation, or live in someone else's cloud.

LLM Limits Tracker is the opposite of that. One file, no dependencies, no signup, your keys never leave your machine. Think of it as glances for LLM APIs.

Features

• 8 providers out of the box — Anthropic, OpenAI, Google Gemini, OpenRouter, Groq, Fireworks, Mistral, Together
• Real rate-limit data — read straight from the providers' response headers (RPM, TPM, input/output tokens)
• Threshold alerts — warning at 80%, critical at 95% (configurable)
• Auto-refresh — optional 60-second polling
• Auto-save settings — paste a key and it's saved + probed on the spot, no Save button
• Three themes — Apple Glass (light & dark) and a green-on-black Hacker mode
• Five languages — English, Русский, Español, 中文, 日本語
• Local-only by default — server binds to 127.0.0.1; keys live in browser localStorage
• Zero dependencies — pure Node.js, no npm install, single binary you can read in 5 minutes

Themes

Quickstart

git clone https://github.com/KovalevAnton/llm-limits-tracker
cd llm-limits-tracker
node server.js

Open http://localhost:5173 and paste any provider's API key into Settings — the dashboard probes it immediately, no Save button, no Refresh.

Or via npm:

npx llm-limits-tracker

To expose it on your LAN (e.g. for a TV dashboard), set HOST=0.0.0.0. The default 127.0.0.1 binding matches the local-only threat model in SECURITY.md.

What you get per provider

| Provider | Metrics | Source | |--------------|--------------------------------------------|---------------------------------------| | Anthropic | RPM, input TPM, output TPM, total TPM | anthropic-ratelimit-* headers | | OpenAI | RPM, TPM | x-ratelimit-* headers | | OpenRouter | Credit balance, RPM (interval-based) | /api/v1/auth/key JSON | | Groq | RPM, TPM | x-ratelimit-* headers | | Fireworks AI | RPM, TPM | x-ratelimit-* headers | | Mistral | RPM, TPM | x-ratelimit-* headers | | Together AI | RPM, TPM | x-ratelimit-* headers | | Gemini | API status (200 / 429 / auth) | response status (no rate-limit headers) |

What it does NOT show

Subscription limits for Claude Pro/Max, ChatGPT Plus/Pro, and Gemini Advanced are not exposed via API — they're closed UI metrics inside the chat products. This tool tracks limits for your API account. A browser-extension companion that scrapes the chat UIs is on the roadmap.

How it works

┌──────────┐   localhost:5173   ┌───────────┐   HTTPS   ┌────────────────┐
│ Browser  │ ◀────────────────▶ │ server.js │ ◀───────▶ │ Provider APIs  │
│ (UI +    │                    │ (proxy +  │           │ (Anthropic,    │
│  state)  │                    │  static)  │           │  OpenAI, ...)  │
└──────────┘                    └───────────┘           └────────────────┘

The proxy exists for one reason: browsers block direct calls to provider APIs (CORS). The proxy forwards them and pipes the response — including rate-limit headers — back to the UI, which parses and renders.

API keys are stored only in your browser's localStorage and travel to the proxy via a single X-Provider-Key header per request. Nothing is persisted on disk by the server.

Roadmap

• [ ] Browser extension for chat-subscription limits (Claude Pro/Max, ChatGPT Plus, Gemini Advanced)
• [ ] Slack / Discord webhook alerts at threshold crossings
• [ ] Persistent history (SQLite) and 7-day usage graphs
• [ ] Per-project tagging — log which app burned which quota
• [ ] More providers — Cohere, Replicate, Perplexity, DeepInfra, Anyscale
• [ ] CLI mode (llm-limits status for terminal use)
• [ ] Docker image for self-hosted setups

See CONTRIBUTING.md for "good first issues".

Where to get API keys

• Anthropic — https://console.anthropic.com/settings/keys
• OpenAI — https://platform.openai.com/api-keys
• Google Gemini — https://aistudio.google.com/app/apikey
• OpenRouter — https://openrouter.ai/keys
• Groq — https://console.groq.com/keys
• Fireworks — https://fireworks.ai/account/api-keys
• Mistral — https://console.mistral.ai/api-keys/
• Together — https://api.together.xyz/settings/api-keys

Project layout

llm-limits-tracker/
├── server.js              # Proxy + static server, zero deps (~210 lines)
├── public/
│   ├── index.html         # Markup only (~115 lines)
│   ├── styles.css         # Three themes, no preprocessor
│   └── app.js             # i18n, providers, parsers, render — vanilla JS
├── docs/                  # Internal notes, launch material (not published to npm)
├── package.json
├── LICENSE                # MIT
├── CHANGELOG.md
├── SECURITY.md
├── CONTRIBUTING.md
└── README.md

Contributing

PRs welcome. The codebase is intentionally small — three short files in public/ and one in server.js, no build step — so contribution friction stays low. See CONTRIBUTING.md for development setup and ideas for first contributions.

License

MIT — do whatever you want, no warranty. See LICENSE.