Run it where
your work belongs.
One computer or a private server. The same clear boundary.
One machine. Your workspace.
For individual work, with the workspace and model running on the computer you use.
- Install the workspace
Run the workspace on your computer and open its local address in your browser.
- Choose a local model
Download a compatible open-weight model. Allow for download size, memory, and disk space.
- Create your first project
Add selected files, set agent permissions, and try a short workflow before adding more work.
Start with a model
your hardware can hold.
Smaller models need less memory and are a practical place to begin. Larger models can need substantially more memory, especially with long documents or concurrent runs.
| Model size | Planning range | Considerations |
|---|---|---|
| 3–4B parameters | 8–12 GB memory | Start small; test short documents first. |
| 7–8B parameters | 16–24 GB memory | Leave room for context and other applications. |
| 13–14B parameters | 24–48 GB memory | Plan more headroom for longer context. |
| 30–32B parameters | 48–64+ GB memory | Assess accelerator memory and speed carefully. |
A few decisions
before you begin.
Allow time for the first download. Model weights can be several gigabytes. Installation and model downloads need network access, even when later inference runs locally.
Start with one run at a time. Increase concurrency after checking memory use. Keeping a model loaded reduces loading time but reserves memory between runs.
Own the server basics. For a private server, arrange authenticated access, encrypted connections, updates, and backups. Keep it on a private network unless you have configured and reviewed remote access.
Talk about your setup