Assistant and models
Assistant is a chat with an AI model. A local model answers on your own computer, and your questions stay there.
Get a local model
Open Marketplace > Models
Each model shows whether it Fits this device or Needs more memory.
Select Get
The download is checked against its signed catalog entry. If it stops, select Retry. To cancel, select Stop.
Open Assistant
Choose the model and start a New chat.
| Model | Size | Notes |
|---|---|---|
| SmolLM2 135M Instruct | about 145 MB | Small and fast. Simple answers. |
| Qwen2.5 1.5B Instruct | about 1.1 GB | Better answers. Needs about 3 GB of free memory. |
If your computer cannot run local AI, Marketplace says Local AI is not supported on this device.
Chat
- Type in Message Assistant and send. Select Stop to stop a reply.
- You can edit or delete a message you just sent, and Regenerate reply.
- Group chats into a New project.
⚠️
Models cannot read your files, open websites or send messages. SmolLM2 135M sometimes says it did one of these things. It did not.
Add a hosted model (Mac, admins)
Hosted models run on another company’s servers. They are optional.
- Open System > Models and, under Hosted, select Add hosted model.
- Choose OpenRouter or Venice and enter your API key.
- Select Check key and load models, choose a model and Save.
Nothing is sent until you approve it in Inbox. To stop it, select End in Inbox, or End hosted approval in System > Models.
Good to know
- Home refuses to start a local model when the computer does not have enough memory for it. Close other Apps or choose a smaller model.
- On a Jetson, getting Qwen takes about 43 minutes and its first reply about 22 seconds.