Super Agentby DigitSpace
Local AI

Your Agent. On your computer.

Super Agent can run an open model on your own computer. Simple steps stay on your machine: private, offline and free to run.

In the app

Two clicks from first launch to a model on your machine.

These are screens from the current build, on a 48 GB Mac.

Super Agent home screen with the “Quick check of your computer” card: this 48 GB Apple-chip computer can run Qwen3-8B on its own, about a 5.0 GB download, with a Download and turn on button and a smaller Qwen3-4B option.
1 · First launch. Super Agent checks your computer and suggests the model that fits, with a smaller option if you prefer.
Settings, Local model: 48 GB memory, 16 cores, Apple silicon, GPU accelerated. Qwen3-0.6B, 1.7B and 4B work; Qwen3-8B is recommended at 5.0 GB and about 40 tokens per second. Deploy local model button.Settings, Local model, in dark mode: the same model list with Qwen3-8B recommended.
2 · Settings › Local model. Every model with its size, speed on your machine and the requests it can take. Deploy, switch or run the check again.
Sized to your machine

It picks the model your computer can run well.

Super Agent looks at your memory and graphics, recommends a model, downloads it, checks the file and runs a quick self-test before it uses it. You can switch or remove it any time in Settings.

8 GB memory
Qwen3-1.7B · 1.8 GB
16 GB memory
Qwen3-4B · 2.5 GB main pick
32 GB+ memory
Qwen3-8B · 5.0 GB

the model left for your other apps · bars drawn to scale

Local first, cloud when it counts

Small jobs stay home. Big jobs get a bigger model.

Short, self-contained steps, such as rewriting a paragraph, pulling dates and amounts out of a text, or sorting a request to the right expert, run on the local model. Larger work goes to the cloud model you choose.

Checked before it is used

Every local answer passes a quality check. If it falls short, the step goes to the cloud model instead, so speed never costs you accuracy.

Nothing leaves your computer

The local model listens on 127.0.0.1 only. Nothing on your network, let alone the internet, can reach it.

Uses what you already run

Already have Ollama, LM Studio or a llama.cpp server? Super Agent finds it on your computer and can use it.

Quiet when you are not using it

The model stops after 10 minutes without work and frees its memory. Downloads are verified before they are used.

Counted as savings

Steps that run locally cost nothing, and they show up in your usage and savings like any other step.

Mac and Windows

Apple silicon uses Metal. On Windows and Linux it uses Vulkan graphics where available, and falls back to the processor.