⚡ Throttle

Universal AI cost meter and model control panel — self-hosted.

Works with every LLM: OpenAI, Anthropic, Google Gemini, Groq, Ollama, LM Studio, and any OpenAI-compatible endpoint.

What it does

• Cost meter — see exactly what you're spending per day, per model, per request
• Cache efficiency — track how much cache is saving you vs. full pricing
• Model switcher — change your active model from a dashboard, no config file editing
• Budget alerts — set a daily limit, get warned before you overshoot
• Request log — every request logged with tokens, cost, latency, and status
• Fully local — no data leaves your machine, no account required

Install

npm install -g throttle-ai
throttle start

• Proxy: http://localhost:4000 — point your agent here
• Dashboard: http://localhost:4001 — opens automatically

How it works

Throttle runs a lightweight proxy between your agent and any LLM provider. You change one URL in your agent's config:

Before:

ANTHROPIC_BASE_URL=https://api.anthropic.com

After:

ANTHROPIC_BASE_URL=http://localhost:4000

Throttle forwards all requests unchanged, logs token counts and costs, and serves a live dashboard.

Supported providers

| Provider | Models | |---|---| | Anthropic | claude-opus-4, claude-sonnet-4-6, claude-haiku-3-5, + more | | OpenAI | gpt-4o, gpt-4o-mini, o3, o4-mini, gpt-4-turbo | | Google | gemini-2.0-flash, gemini-1.5-pro, gemini-1.5-flash | | Groq | llama-3.3-70b, llama-3.1-8b, mixtral-8x7b | | Ollama | Any locally running model | | LM Studio | Any locally running model | | OpenRouter | Any model via OpenRouter | | Custom | Any OpenAI-compatible endpoint |

Data & privacy

All data is stored locally in ~/.throttle/throttle.db (SQLite). Nothing is sent anywhere. No account, no telemetry, no cloud.

License

MIT — use it, fork it, ship it.