Ollama: MLX Threading Fixes and Claude App Integration

Two significant updates merged today: threading fixes for MLX on macOS addressing OS-thread-local execution state, and full Claude App integration with launch commands.

Duration: PT1M35S

Episode overview

This episode is a short developer briefing from Ollama.

It explains recent repository work in plain language.

  • Show: Ollama
  • Published: 2026-05-03T00:00:00Z
  • Audio duration: PT1M35S

Transcript excerpt

This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.

Good morning, this is your Ollama developer briefing for Saturday, May 3rd, 2026.

Daniel Hiltgen merged a critical MLX threading fix addressing OS-thread-local execution state issues on macOS. MLX now requires streams, encoders, and caches to remain bound to specific operating system threads. The update introduces a new mlxthread executor using Go's runtime.LockOSThread to prevent goroutine…

Parth Sareen merged full Claude App integration, enabling users to launch Claude directly through Ollama with `ollama launch claude-app`. The integration supports Claude Cowork and Code features within the app, with a restore option available through the `--restore` flag. Launch settings persist between sessions.…

Both changes include corresponding commits with the same content as their pull requests.

What's next: The MLX threading fixes should resolve stability issues for macOS users running Apple Silicon inference workloads. The Claude integration expands Ollama's ecosystem connectivity with major AI applications.

That's your Ollama update for today. We'll be back Monday with more developer news.

Nearby episodes from Ollama

  1. Launch Command Enhancements
  2. Speed Revolution - MTP Decoding and Smart Caching
  3. Go 1.26 Runtime Update
  4. Weekly Recap - MLX Threading & Model Recommendations
  5. Model Recommendations and Windows Gateway Fix
  6. Metal GPU Stability and Gemma4 Updates
  7. Launch Experience Improvements and Model Recommendations
  8. Multi-Sequence Batching and New Model Support