Ollama: Startup Performance Optimization

Transcript

Good morning, this is your Ollama development briefing for Monday, May 20th, 2026.

Daniel Hiltgen merged a significant performance optimization in pull request 16215, titled "Reduce startup model hydration." This change addresses issue 16123 by implementing a lightweight model list cache for tags and launch inventory. Previously, Ollama would load every local model at startup, creating delays for users with large model stores. The new implementation keeps show cache population lazy, meaning models are only fully loaded when needed. This substantial change spans 16 files with over 1,300 lines added, including new cache management code and comprehensive test coverage.

Bruce MacDonald merged a smaller but important compatibility fix in pull request 16231. The change removes a patch tool type from the codex implementation that was causing schema compatibility issues. This aligns with recent updates to the codex specification.

Both changes were also reflected in additional commits to the main branch, confirming the merges are now live in the codebase.

What's next: These startup performance improvements should significantly benefit users managing multiple models locally. The codex compatibility fix ensures smoother integration with external tools and services.

That's your Ollama development update. We'll be back tomorrow with more changes from the team.