Local Model + OpenRouter

[ShailChoksi]

Maybe I am not understanding your .env correctly but can I do the following:

• have a local model running on localhost
• and also have OpenRouter setup for larger models?

Currently I only see 1 env var that handles the URL for both is that true?

[nonlooped]

I don't believe you can currently do that. Fast/Default/Strong can route to different model names but they all share the same base URL so you cannot cleanly use localhost for one tier and OpenRouter for another.

[ShailChoksi]

Yes - I figured.

It would be nice to have it as a feature then :)

[Doorman11991]

Fixed in the next release. You can now mix a local model with a cloud endpoint using per-tier base URLs.

Add to your .env:
\\�ash

Default tier — local model

SMALLCODE_MODEL=qwen3-8b
SMALLCODE_BASE_URL=http://localhost:11434/v1

Strong tier — OpenRouter (used automatically when local model fails repeatedly)

SMALLCODE_MODEL_STRONG=openai/gpt-4o-mini
SMALLCODE_BASE_URL_STRONG=https://openrouter.ai/api/v1
OPENROUTER_API_KEY=sk-or-v1-...
\\

The adaptive router switches to the strong model when the local model's failure rate exceeds 60% (after at least 3 calls). You can also trigger it manually by setting SMALLCODE_MODEL_STRONG — it will be used for escalation when the local model hard-fails.

The new SMALLCODE_BASE_URL_STRONG and SMALLCODE_BASE_URL_MEDIUM env vars are documented in .env.example.