v1.5.1

What's Changed

Bug Fixes

• Gemini reasoning-only models no longer hang in headless mode: When using Gemini 3.x or other reasoning-heavy models on the direct Google provider path, the conversation mirror now properly captures reasoning-only output that has no separate text part. Previously these models would timeout while still consuming tokens because the output was never detected. The fix accumulates reasoning parts separately and promotes them to the final output only when no visible text part arrives, ensuring models that emit both reasoning and text remain unaffected.

Full Changelog: v1.5.0...v1.5.1