Frequently Asked Questions
What problem does Claude Code Router solve?
It solves the growing friction between AI coding utility and operational sustainability—eliminating the trade-off between model capability and cost. Instead of choosing *one* LLM and accepting its limitations, developers now orchestrate *many*, letting each model do what it does best—guided by rules they define, not algorithms they can’t audit.
How is this different from just calling multiple APIs manually?
Manual orchestration means writing glue code, handling auth per provider, managing failures, normalizing responses, and maintaining state across sessions. Claude Code Router abstracts all of that—providing a consistent CLI interface, shared history, unified configuration, and intelligent fallback logic—so developers focus on *what to build*, not *how to route*.
Which models and providers are supported out of the box?
Out-of-the-box support includes OpenRouter (access to >100 models), DeepSeek-Coder, Ollama (Llama 3, Phi-4, CodeLlama), Google Gemini (1.5 Flash/Pro), and Volcengine’s Doubao series. Thanks to its provider-agnostic adapter pattern, adding new backends—like Groq, Fireworks, or self-hosted vLLM instances—requires only a few lines of JSON and optional TypeScript extensions.
Is configuration mandatory—or does it work “out of the box”?
Yes—it ships with a sensible default config that enables immediate use with Anthropic’s Claude models. But true power unlocks when you customize: define preferred fallbacks, disable expensive models in staging environments, or restrict certain providers to specific branches via CI-environment variables. All without touching source code.
Can I override routing decisions during a live session?
Absolutely. Use /model to change providers instantly, /role think to force high-reasoning mode, or /cost max=0.05 to cap per-request spend—even mid-conversation. Every command updates the active routing policy in real time, giving you granular, interactive control over AI behavior.