Add configurable base URL for OpenAI-compatible providers

[rubenvanerk]

Summary

Adds support for using OpenRouter and other OpenAI-compatible API providers by allowing users to configure a custom base URL via the LLM_BASE_URL environment variable.

Why: OpenRouter provides rate limiting and cost controls per API key, which is useful for managing expenses across different projects or workflows.

Changes

• Add LLM_BASE_URL config option (environment variable or GitHub Action input)
• Pass baseURL to AI SDK provider when configured
• Skip model whitelist validation when custom base URL is set, enabling use of provider-specific model names (e.g., openai/gpt-4o-mini)
• Add unit tests for configuration loading
• Update documentation with usage examples

Implementation Notes

When a custom LLM_BASE_URL is provided, the OpenAI SDK is used regardless of the model name. This works for OpenRouter and other OpenAI-compatible providers, but users should be aware of this assumption.

Note

This implementation bypasses the model whitelist when using a custom base URL. This workaround may become unnecessary after PR #43 is merged, which refactors the model configuration system.

Tested here: rubenvanerk#1

[github-actions]

PR Summary

Added support for configurable base URLs to enable OpenAI-compatible API providers (e.g., OpenRouter, Anyscale, Together AI) via the LLM_BASE_URL environment variable or GitHub Action input. When a custom base URL is provided, the OpenAI SDK is used and model whitelist validation is bypassed to allow provider-specific model names. This feature only applies to the ai-sdk provider and includes comprehensive unit tests and documentation updates.

Changes

File	Summary
.env.example	Added optional LLM_BASE_URL environment variable example with comment explaining it's for OpenAI-compatible providers when using LLM_PROVIDER=ai-sdk. Includes example URL for OpenRouter.
README.md	Added LLM_BASE_URL to required configuration list and created new section explaining how to use OpenRouter and other OpenAI-compatible providers. Includes YAML configuration example and notes that this feature only works with ai-sdk provider. Also added LLM_BASE_URL to local development setup instructions.
dist/cli.js	Rebuilt distribution file to include the new LLM_BASE_URL configuration changes.
dist/index.js	Rebuilt distribution file to include the new LLM_BASE_URL configuration changes.
src/__tests__/config.test.ts	Added three test cases: loading LLM_BASE_URL from environment variable, verifying it's undefined when not set, and loading it from GitHub Action input. Tests cover both configuration sources and default behavior.
src/ai.ts	Modified runPrompt function to skip model whitelist validation when config.llmBaseUrl is set and provider is ai-sdk. Creates a fallback modelConfig using createOpenAI to support provider-specific model names when custom base URL is configured.
src/config.ts	Added llmBaseUrl property to Config class, loading from LLM_BASE_URL environment variable or llm_base_url action input. Updated mock config to include llmBaseUrl: undefined for test environments.
src/providers/ai-sdk.ts	Modified AISDKProvider.infer method to conditionally pass baseURL option to createAiFunc when config.llmBaseUrl is set, enabling custom API endpoints for OpenAI-compatible providers.

---

autogenerated by presubmit.ai

[shawnd]

@presubmit

[bstanga]

Thanks!