# BurnRate Proxy — Usage Tracking for AI API Calls

BurnRate is a macOS menu bar app that tracks AI token usage. It runs a transparent local proxy that logs every API request's token count without modifying requests or responses.

## What to do

Configure your AI provider base URLs to route through BurnRate's local proxy. **API keys stay the same** — BurnRate just observes the traffic.

### Provider Base URLs

| Provider | Default URL | BurnRate Proxy URL |
|----------|-------------|------------------|
| Anthropic (Claude) | `https://api.anthropic.com` | `http://localhost:8462/anthropic` |
| OpenAI | `https://api.openai.com` | `http://localhost:8462/openai` |
| Kimi / Moonshot | `https://api.moonshot.ai` | `http://localhost:8462/kimi` |
| MiniMax | `https://api.minimax.chat` | `http://localhost:8462/minimax` |

### For OpenClaw gateway config

In your gateway's provider configuration, replace each provider's `baseUrl` with the corresponding proxy URL above. Example for Anthropic:

```json
{
  "providers": {
    "anthropic": {
      "baseUrl": "http://localhost:8462/anthropic",
      "apiKey": "YOUR_EXISTING_KEY"
    }
  }
}
```

Repeat for each provider you want to track. Model names, API keys, and all other settings stay the same.

### For CLI tools (environment variables)

```bash
# Add to ~/.zshrc or ~/.bashrc
export ANTHROPIC_BASE_URL=http://localhost:8462/anthropic
export OPENAI_BASE_URL=http://localhost:8462/openai
```

## How it works

```
Your tool → localhost:8462/anthropic/v1/messages → api.anthropic.com/v1/messages
                         ↓
                  BurnRate logs token usage
                         ↓
              Response returned unmodified
```

- BurnRate must be running (antenna icon in macOS menu bar)
- Listens on `localhost:8462`
- API keys pass through transparently and are never stored
- Works with both streaming and non-streaming responses

## Verify

```bash
curl http://localhost:8462/anthropic/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model":"claude-sonnet-4-6","max_tokens":10,"messages":[{"role":"user","content":"Hi"}]}'
```

A normal Claude response means the proxy is working.