Skip to content

fix: bypass gateway for local model connection tests#472

Open
Bortlesboat wants to merge 1 commit intorowboatlabs:mainfrom
Bortlesboat:codex/rowboat-local-model-test-routing
Open

fix: bypass gateway for local model connection tests#472
Bortlesboat wants to merge 1 commit intorowboatlabs:mainfrom
Bortlesboat:codex/rowboat-local-model-test-routing

Conversation

@Bortlesboat
Copy link
Copy Markdown

Summary

Fix issue #471 by making local provider connection tests bypass the Rowboat gateway even when the user is signed in.

Today testModelConnection() routes all signed-in tests through getGatewayProvider(). That works for Rowboat-managed hosted models, but it breaks local providers because the gateway doesn't know local model ids like llama3.3:70b or qwen3.5:latest.

This patch keeps the current gateway behavior for hosted providers while always testing these local providers directly:

  • ollama
  • openai-compatible

Changes

  • add shouldUseGatewayForModelTest() in apps/x/packages/core/src/models/models.ts
  • update testModelConnection() to bypass the gateway for signed-in local providers
  • add a focused regression test in apps/x/packages/core/src/models/models.test.ts

Testing

  • pnpm test in apps/x/packages/core
    • verifies signed-in ollama bypasses the gateway
    • verifies signed-in openai-compatible bypasses the gateway
    • verifies hosted providers still use the gateway when signed in

Notes

I kept this intentionally narrow to the failing connection-test path. I did not change onboarding model-list behavior because the local-provider UI already falls back to manual model entry.

Fixes #471.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Desktop app onboarding: "Model not allowed" for all Ollama models (v0.2.0, macOS)

1 participant