Currently anthropic's claude sonnet is hardcoded into the codebase. I would like to use different LLM provided like OpenAI or openrouter or even LiteLLM proxy. I checked through the vercel's workflow library documentation but it doesn't provide a straight forward example to use our own OpenAI compatible configuration so that we can provide it with a base URL and API keys. Would be appreciated if this is documented somewhere
Activity
2
amine-y7mo agoCurrently anthropic's claude sonnet is hardcoded into the codebase. I would like to use different LLM provided like OpenAI or openrouter or even LiteLLM proxy. I checked through the vercel's workflow library documentation but it doesn't provide a straight forward example to use our own OpenAI compatible configuration so that we can provide it with a base URL and API keys. Would be appreciated if this is documented somewhere
@shivam-clary Totally agree
This is exactly the concern we had when designing our pipeline. Provider lock-in is a real risk — if Anthropic changes pricing, goes down, or deprecates an API version, your entire review flow breaks overnight.
Beyond provider flexibility, we also found that some AI-code defects can be caught deterministically without needing an LLM at all: hallucinated package imports (referencing packages that don't exist on the registry), deprecated API calls that still work in pinned dev versions but break in production, and cross-file inconsistencies from multi-session AI generation.
We combine both approaches: deterministic checks for known patterns + optional AI analysis for nuanced review. It makes the pipeline faster, cheaper, and more resilient to provider changes.
For the deterministic part:
npx @opencodereview/cli scan . --sla L1