🛂 Tollgate

A local proxy that intercepts your Anthropic API traffic, monitors token spend, and optionally routes to cheaper models before your budget runs out.

You're burning through your Claude Max budget and you have no idea until it's too late.

Tollgate fixes that. It's a zero-config local proxy that sits between your AI tools and Anthropic's API, reads the rate limit headers on every response, and shows you exactly how close you are to the wall — before you hit it.

Works with Claude Code, Cursor, Continue.dev, Aider, raw API scripts — anything that uses ANTHROPIC_BASE_URL. No code changes. No account. No data leaves your machine.

Quickstart

npx tollgate start

Then point your tool at the proxy:

export ANTHROPIC_BASE_URL=http://localhost:4243

For Claude Code, add to ~/.claude/settings.json:

{ "env": { "ANTHROPIC_BASE_URL": "http://localhost:4243" } }

Open your dashboard: http://localhost:4244 🛂

That's it. Tollgate is open for business.

What You Get

| Feature | Details | |---------|---------| | 📊 Live token gauge | Big visual showing % of budget used, color shifts green→red | | ⏱️ Reset countdown | Exact time until your rate limit window refreshes | | 💰 Cost per call | Real $ estimated from Anthropic's pricing per model | | 🔍 Live call feed | Every API call logged: model, tokens in/out, cost, latency | | 📈 Burn rate | Tokens/min so you can project when you'll hit the wall | | 🚦 Smart routing | Auto-downgrades Sonnet→Haiku at 80% budget (opt-in) | | 🔔 Push alerts | Browser notifications before you run out | | 💾 Full history | SQLite log of every call — query it yourself |

Smart Routing (opt-in)

When you're approaching your limit, Tollgate can automatically downgrade your model requests to save budget:

tollgate start --routing

🚦 Rerouted: sonnet → haiku (tokens at 81%) — budget preserved

The downgrade ladder: opus → sonnet → haiku. Transparent to your client — we just swap the model field before forwarding.

Why Not LangSmith / Helicone / AgentOps?

| | Tollgate | Others | |--|----------|--------| | Local-first | ✅ Your machine only | ❌ Cloud/SaaS | | Zero code changes | ✅ One env var | ❌ Requires SDK | | Works with any tool | ✅ Proxy-based | ❌ Per-SDK instrumentation | | Rate-limit focus | ✅ First-class | ❌ Afterthought | | Free & open source | ✅ MIT | ❌ Paid tiers |

Dashboard

(screenshot coming soon)

Install Globally

npm install -g tollgate
tollgate start

CLI Options

tollgate start              Start proxy + dashboard
tollgate start --routing    Enable smart model routing at 80% budget
tollgate start --port 4243  Custom proxy port (default: 4243)
tollgate start --dashboard-port 4244  Custom dashboard port (default: 4244)

Configuration

Config lives at ~/.config/tollgate/config.json:

{
  "proxy": { "port": 4243 },
  "dashboard": { "port": 4244, "autoOpen": true },
  "routing": {
    "enabled": false,
    "threshold": 80,
    "ladder": {
      "claude-opus-4-6": "claude-sonnet-4-6",
      "claude-sonnet-4-6": "claude-haiku-4-6"
    }
  },
  "alerts": {
    "tokenWarningPercent": 80,
    "tokenCriticalPercent": 95
  }
}

Contributing

See CONTRIBUTING.md. The one rule: don't break the zero-config promise.

License

MIT © AlexandeCo