Ollama: The MLX Engine Comes Home
The MLX runner graduated out of experimental status into a full top-level engine with a seven-commit cleanup pass, while a cluster of parser and schema fixes tightened up tool-calling and thinking behavior across the API.
Duration: PT2M42S
Episode overview
This episode is a short developer briefing from Ollama.
It explains recent repository work in plain language.
- Show: Ollama
- Published: 2026-09-17T13:02:36Z
- Audio duration: PT2M42S
Transcript excerpt
This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.
Good morning. It's September 17th, 2026, and this is Ollama's daily briefing.
The headline today is a graduation: the MLX engine is no longer experimental, and Jesse Gross spent seven commits and PR 18489 moving it out of the x folder for good.
That move is bigger than a rename. The bindings now live in a top-level MLX package mirroring how llama-dot-c-p-p is organized, the runner becomes "mlx runner," and along the way, dead code got pulled out by the roots. Commit bef41f7 removed the old Go engine's backend interfaces, an unused CUDA template, and a…
The second theme is correctness in how Ollama talks to models and tools. Three separate fixes landed for the same category of bug: something works when you don't ask for a feature, but breaks when you explicitly request it. PR 18504 fixes reasoning parsing on the generate endpoint when thinking is explicitly turned…
Smaller but worth flagging: PR 18508 adds Prometheus metrics support so servers can finally track load and generation time over time, and PR 18500 stops orphaned runner processes on Windows by tying them to a job object that Windows can force-kill.
What's next: watch for the MLX directory move rippling into any tooling that…
Nearby episodes from Ollama
- Streaming Reliability and Parser Cleanup
- MLX Comes of Age, Cancellation Gets Fixed
- Consistency Fixes for Tool Schemas and Image Input
- Weekly Recap - Memory Discipline and Tool-Calling Cleanup
- Precision Fixes and Cleanup
- Silent Failures Get Loud
- Taming Memory in the MLX Runner
- Tool-Call Parsing Gets a Hardening Pass