OpenClaude Router (OpenRouter-first)

OpenClaude Router keeps the Claude Code agent/tool runtime, but routes model traffic through an OpenRouter-first OpenAI-compatible shim.

Quick start

export CLAUDE_CODE_USE_OPENAI=1
export OPENROUTER_API_KEY=sk-or-...
export OPENROUTER_MODEL=openai/gpt-4o-mini
# optional
export OPENROUTER_BASE_URL=https://openrouter.ai/api/v1
export OPENROUTER_STREAM=auto   # auto|on|off
export OPENROUTER_MODELS=openai/gpt-4o-mini,google/gemini-2.5-flash

openclaude

Compatibility aliases supported in this PR:

• OPENAI_API_KEY → OPENROUTER_API_KEY
• OPENAI_BASE_URL → OPENROUTER_BASE_URL
• OPENAI_MODEL → OPENROUTER_MODEL

Profiles

Set OPENROUTER_PROFILE or use launcher scripts:

• free (opt-in only; never default for autonomous coding)
• balanced
• reliable
• cheap

Commands:

bun run dev:openrouter
bun run dev:free
bun run dev:cheap
bun run dev:reliable

Behavior notes

• If selected model metadata indicates weak/no tools support, runtime warns and degrades by omitting tool declarations rather than failing deep in the loop.
• Model fallback chain: OPENROUTER_MODEL + optional OPENROUTER_MODELS + profile defaults.
• OpenRouter /models metadata is cached at .openclaude/openrouter-models-cache.json.
• Streaming mode is controlled via OPENROUTER_STREAM=auto|on|off.

Runtime checks

bun run doctor:runtime
bun run doctor:runtime:json
bun run doctor:report

These validate OpenRouter config, compatibility aliases, reachability, and local-provider constraints.