builders · review · first seen Oct 27, 09:33 UTC

Unsloth adds scheduled tasks and Mac performance updates

Unsloth shipped scheduled prompts, per-model Run presets and a Mac chat performance update. The decision for builders is whether prompts that run only while Unsloth is open, without tools or web search, fit the work they want to schedule.

L0 · primary1 source5 claims checked4 min read

What shipped

Scheduled tasks can run a prompt once, daily, on weekdays, weekly or monthly. Open Scheduled from the sidebar’s More menu to set one up. Each answer is saved as a new chat. That gives a builder a place to inspect individual results, but the release does not describe a combined report or another destination for those answers. [1]

A task can use the model already loaded, another model on the device, or a model from a connected API provider. If its model is not loaded, Unsloth asks before loading it unless the user enables the option to switch automatically while away. That choice matters before leaving a task unattended: a prompt and a schedule alone do not settle how an unloaded model will be handled. [1]

Tasks run only while Unsloth is running. They cannot use tools or web search, so the shipped feature is scheduled prompting rather than a way to run a tool-using workflow later. The release also adds named Run-setting presets for each model, with options to save, rename, delete and switch between them. Separately, MLX chats with draft models now reply up to 16% faster on Mac. The stated gain applies to that chat setup, not to every Mac task. [1][2][3]

Why it matters if you build

The most direct fit is a prompt a builder already knows how to run with an available model and wants to repeat on a stated cadence. Saving each answer as a new chat gives that workflow an inspectable result. It does not remove the need to check those chats or decide what to do with their contents. Start with one prompt whose output can be judged on its own, rather than treating the schedule as an end-to-end process. [1][2]

Availability is the operational limit. If Unsloth is closed, the task does not run. If the selected model is not loaded, the default behavior described in the release is to ask before loading it; automatic switching is an option the user must enable. Builders who need a run without an open Unsloth session, or who cannot accommodate that model-loading decision, should not assume this release covers their case. The supplied material does not say how a missed run is handled. [1][2]

Presets address a different source of friction: keeping named Run settings with each model and switching among them. That may help when comparing repeatable configurations, but the release does not establish that a scheduled task selects a preset. Mac users should also keep the performance statement narrow. The reported improvement is up to 16% for MLX chats with draft models; it is not a measured reduction in scheduled-task runtime or a guarantee for a particular machine. [1][2][3]

How to try it today

1. Open Scheduled from the sidebar’s More menu.
2. Set one prompt to run once, then inspect its answer in the new chat.
3. Choose the loaded model, a model on the device, or a connected API provider.
4. For an unloaded model, decide whether to allow automatic switching while away.
5. Choose a notification setting: every run, failures only or never.
6. If useful, save named Run settings in the model picker and switch between them.

What we checked

Every claim below was walked down the source ladder. The rung is how close the source is to the thing itself; the tense is the source's own.

1. shipped L0 · primary

   Unsloth adds scheduled tasks that can run prompts once, daily, on weekdays, weekly or monthly.

   per GitHub releases · unslothai/unsloth (L0)
2. shipped L0 · primary

   Scheduled tasks can use a loaded model, a model on the device, or a connected API provider.

   per GitHub releases · unslothai/unsloth (L0)
3. shipped L0 · primary

   Scheduled tasks run only while Unsloth is running and cannot use tools or web search.

   per GitHub releases · unslothai/unsloth (L0)
4. shipped L0 · primary

   Unsloth adds named Run-setting presets for each model.

   per GitHub releases · unslothai/unsloth (L0)
5. shipped L0 · primary

   MLX chats with draft models now reply up to 16% faster on Mac.

   per GitHub releases · unslothai/unsloth (L0)

What is not confirmed

The supplied release excerpt does not establish what happens to a scheduled occurrence when Unsloth is not running, or whether it runs later after Unsloth opens. It establishes only the condition for running: Unsloth must be running. Do not build a time-critical process around an assumed catch-up behavior. [1]

The supplied material does not establish scheduled-task pricing for a connected API provider, a performance gain for scheduled tasks, or a way for scheduled prompts to use tools or web search. It also does not say that named Run presets are used by scheduled tasks. Those are separate questions to resolve before relying on any of those behaviors. [1][2][3][4]

Who should act, who should wait

Act now if you want a repeatable prompt, can keep Unsloth running for the scheduled time, and can work with an answer saved as a new chat. Test a single run with the intended model first. Then choose a recurring cadence and notification setting that let you review the results. [1][2][3]

Wait if the workflow depends on Unsloth running while closed, tool calls, web search, or a confirmed way to recover missed runs. If the reason to move is Mac performance alone, test the specific MLX chat setup with a draft model rather than applying the up-to-16% figure to other work. [1][2]

Sources

• Scheduled + MiniMax-M3.1-Flash + Perf Gains — GitHub releases · unslothai/unsloth L0 · primary