Summary

The onboarding flow still presents local-model options even though local model support was descoped from the current release.

Problem

Users are prompted to choose between local models and API endpoints, and Ollama may be auto-selected if detected, even though the intended reduced-scope release is focused on inference endpoints. This creates a misleading and sometimes non-functional setup path.

Expected behavior

Only supported inference options should appear in onboarding. Unsupported local-model flows should be hidden behind a feature flag or removed until they are ready.

Acceptance criteria

• Onboarding only presents inference options that are supported in the current release.
• Existing local-model code paths are hidden or gated behind an explicit experimental flag.
• The installer does not auto-configure a local provider without user confirmation.
• Docs and release notes match the actual supported onboarding flow.