Ollama: Adaptive Memory on Macs, Tighter Server Security

Merged work lets Metal runners scale saved conversation history to real memory pressure instead of a fixed allowance, with a related proposal to auto-size parallelism. The rest adds new model support, a server hardening fix, and power and ecosystem docs.

Duration: PT2M26S

Episode overview

This episode is a short developer briefing from Ollama.

It explains recent repository work in plain language.

  • Show: Ollama
  • Published: 2026-11-20T14:02:10Z
  • Audio duration: PT2M26S

Transcript excerpt

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

Hello, it's Friday, November 20th, 2026.

The lead story is Ollama getting smarter about memory on Apple Silicon — keeping more work when memory is free, and letting go when it isn't.

First theme is adaptive memory in the Metal runner. The merged change in PR 19678, commit 73ee71d, replaces a fixed 8 gigabyte allowance for saved conversation history. Instead it looks at observed headroom, execution workspace, and system compression, retaining reusable conversation prefixes when possible and…

Second, capability and hardening. One open change, PR 19783, ports text and vision support for Hy 4 Flash to MLX, along with a new speculative drafting adapter that conditions proposals on the target cache. Separately, PR 19786 would serve the API on its own handler instead of the default web router. Today nothing…

Finally, docs and ecosystem. PR 19788 notes in the keep alive FAQ that a loaded model keeps the GPU awake — about 58 watts with a 7 billion parameter model loaded on an A-10 test, versus 16 watts idle. And four community updates add tools like Voxmelt, Athena and Karpa, and refresh Big-AGI.

What's next to watch is whether the auto-parallelism and Hy 4 work merge, and whether the server…

Nearby episodes from Ollama

  1. Inkling on Mac Gains and Stream Reliability Fixes
  2. Tool Call Reliability and Runner Efficiency
  3. Runner Reuse, Decision Models, and Thinking UI Fixes
  4. MLX Concurrency and Memory Tuning
  5. GPU Load Fixes and Clearer Errors
  6. Weekly Recap - Parallel Inference, Tool Fidelity, and Launcher Polish
  7. Qwen Tool-Call Reliability Fix
  8. Tool Schema Fidelity, Desktop Launch, and Download Safety