Thrum

What is modelmanagerd, and why is it using so much memory?

Last reviewed: 19 November 2026. Path, account, launchd job and man page checked on macOS 26.3.1.

modelmanagerd is the macOS Model Management Daemon: it manages machine learning models and the requests to run them. Its memory use rises when models are loaded for features that need them and falls when they are released, so large numbers while you use AI features are expected. It is part of macOS and runs as its own system account.

modelmanagerd at a glance

modelmanagerd
Made byApple (part of macOS)
TypeSystem daemon
Location/usr/libexec/modelmanagerd
Runs as_modelmanagerd (a system account)
Started bylaunchd job com.apple.modelmanagerd
Can you quit it?Not usefully. launchd starts it again when models are requested.

What is modelmanagerd?

The modelmanagerd(8) man page says it manages ML models and requests for execution of those models, has no options, and should not be run manually. It runs as the system account _modelmanagerd from the com.apple.modelmanagerd launchd job and offers a service named com.apple.modelmanager.

Because it handles requests to execute models, memory used for those models can show up under modelmanagerd rather than under the app that asked. The catalog of available models is kept by modelcatalogd.

Is modelmanagerd safe?

Yes. modelmanagerd is part of macOS at /usr/libexec/modelmanagerd and runs as a restricted system account.

Why does modelmanagerd use so much memory and CPU?

How much memory is normal for modelmanagerd?

Typical: modest memory when no model is in use, and much more while models are loaded.

A large memory figure while you are using AI features is expected; constant high memory pressure with no such use is not.

What can you do about modelmanagerd's memory use?

  1. Check memory pressure, not just the number. Open Activity Monitor's Memory tab and look at the Memory Pressure graph. If it is green, the memory modelmanagerd holds is not slowing the Mac.
  2. Close apps using AI features. Finish using summaries, writing tools or image features; models can then be released.
  3. Turn off Apple Intelligence if you do not use it. System Settings > Apple Intelligence & Siri lets you turn Apple Intelligence off, which removes the features that need its models.
  4. Restart the Mac. If its memory stays high for a long time with nothing using models, restart.

When should you worry about modelmanagerd?

Worry if modelmanagerd holds a lot of memory while memory pressure is yellow or red and you are not using any AI feature, or if it crashes repeatedly.

A modelmanagerd outside /usr/libexec would not be Apple's.

How do you check modelmanagerd in Activity Monitor and Terminal?

Open Activity Monitor (Applications > Utilities), choose View > All Processes, and type modelmanagerd in the search field. The Memory tab shows its Memory. The CPU tab shows its % CPU. Double-click the row for its parent process, memory details and open files.

In Terminal, the first command lists every modelmanagerd process with its PID, user, CPU and memory (RSS is resident memory in kilobytes).

ps -axo pid,user,%cpu,%mem,rss,command | grep -F 'modelmanagerd' | grep -v grep

How can Thrum help with modelmanagerd?

Thrum's Memory tab shows memory pressure, swap, and the eight processes using the most memory, so you can see how much modelmanagerd holds and whether the Mac is short of memory. An optional alert tells you when memory pressure turns critical. macOS does not let apps quit processes owned by root or other users, so Thrum shows them for information and leaves stopping them to the steps above.

Thrum is $4.99 once with a 14-day free trial and no account. It requires macOS 26 Tahoe on Apple silicon or Intel.

Questions about modelmanagerd

Why does modelmanagerd use gigabytes of memory?

It loads machine learning models to run them, and large models need a lot of memory. The memory is released when the models are no longer needed.

What is the difference between modelmanagerd and modelcatalogd?

modelcatalogd provides the catalog of models, adapters and tokenizers; modelmanagerd manages models and runs them on request.

Sources