A lightweight macOS menu bar proxy that intelligently routes LLM API requests across 42+ providers with automatic failover, protocol conversion, and real-time usage analytics — all running locally on your Mac.
SmartLLMRouter replaces multiple tools with a single, lightweight native app.
Priority-based routing with intelligent cooldown — handles 429/5xx/401 errors silently. Three-layer fallback ensures your request never fails.
Seamless conversion between Anthropic and OpenAI formats. Use Claude Code with OpenAI-compatible providers, or vice versa.
Track daily token usage and estimated costs with 30-day history. Know exactly where your API budget goes.
100% local execution. API keys stored in macOS Keychain. No telemetry, no cloud sync, no third-party servers.
Pre-configured metadata for DeepSeek, OpenAI, Anthropic, Aliyun, MiniMax, and more. One-click setup via providers.json.
Just set ANTHROPIC_BASE_URL. Claude Code and other clients work out of the box — no plugins, no wrappers, no config files.
Export all channels with optional encryption and share with colleagues. One-click import — no manual re-entry. API keys encrypted, never visible.
See how SmartLLMRouter compares to other solutions.
| SmartLLMRouter | Cloud Proxies | Other OSS | |
|---|---|---|---|
| Binary Size | ~8 MB | 100+ MB | 50+ MB |
| macOS Native | ✓ SwiftUI | Web UI | Electron/CLI |
| Privacy | 100% Local | Cloud Relayed | Partial |
| Auto-Failover | 3-Layer | Limited | Manual |
| Protocol Conversion | ✓ Built-in | Partial | No |
| Key Storage | Keychain | Server-side | Config file |
| Setup Time | < 1 min | 10+ min | 5+ min |
Use multiple API keys for the same model. When DeepSeek hits rate limits, the proxy silently switches to Nvidia or Sensenova — same model, different provider.
export ANTHROPIC_BASE_URL=http://localhost:1897
claudeWhen a provider returns 429, the circuit breaker trips and the request automatically retries with the next priority provider. You see no interruption.
When a request exceeds a model's context window, SmartLLMRouter intelligently routes to a compatible model with larger capacity — like from gpt-4o (128K) to gpt-4.1 (1M).
Run local models (Ollama, vLLM) alongside cloud APIs. SmartLLMRouter unifies them behind a single endpoint — one URL serves all models.
Free · Open Source · macOS 13+ · Apple Silicon & Intel