Castellan
Overview · Smith
The Smith, which keeps your local AI ready
What the Smith does for the other agents' AI work, what ready, busy, resting and not working mean on its page, how it frees your PC's memory, and its settings.
- Article
- 1401
- Applies to
- Smith 0.8.3
- Last reviewed
- For
- For everyone
What the Smith does#
The other agents use a small AI model on this PC for their AI work: to describe a picture, label a log, or search by meaning. The model runs on your PC's NPU, a graphics card, or the processor (see Where the work runs). The Smith looks after it:
- It sets the local AI up, on the accelerators you choose. See Set up the local AI.
- It keeps it working. Once a minute it looks at the local AI on each accelerator, and starts it again if it stops answering.
- It lets it rest. When nobody has used an accelerator for 10 minutes, the Smith stops the local AI there, so its memory is yours again. It starts again by itself, in a few seconds, as soon as an agent needs it.
- It makes way for games. Once a game is using a graphics card and nobody else has for 2 minutes, the local AI there stops.
The Smith comes with the Household license.
Its page#
The Smith's page is at http://smith.localhost:20202/, and Open on its card in Castellan goes there.
The local AI shows each of your accelerators (the NPU, each graphics card, the processor) with how its local AI is doing:
| It says | What it means |
|---|---|
| ready | Running, and ready for your agents. |
| busy | Starting, or answering one of your agents. |
| resting | Stopped to give your memory back. It starts again by itself as soon as an agent needs it. |
| not working | It couldn't start. Your agents use another accelerator meanwhile, and it's tried again within 10 minutes. |
| not set up | No local AI here yet, so no agent can use it. See Set up the local AI. |
Each accelerator's card shows how much of it is in use and who's waiting, and Switch off stops your agents using it until you Switch on again.
What it did lists what the Smith did lately, newest first: when it let the local AI rest, or woke it again.
Run now, in the title bar, has the Smith look at once rather than at its next minute.
Its settings#
Settings, in the title bar:
| Setting | Default | What it does |
|---|---|---|
| Let the local AI on the NPU rest after | 10 minutes | How long nobody uses the NPU before the local AI there stops. 0: never. |
| Let the local AI on a graphics card rest after | 10 minutes | The same for a graphics card or the processor. A game using the card stops it after 2 minutes, whatever this says. 0: never. |
| The local AI on your NPU (under Advanced) | The newest that works here | Which setup the NPU uses: the newest that works, or the one it had before the last change. Choosing the one before puts it back. |
What it never does#
- It never touches an AI server it didn't set up. One you run yourself, or another program's, is left alone.
- It never takes an agent's turn. When it stops or restarts the local AI, it waits until nobody is using that accelerator.
- Nothing leaves your PC. The local AI runs here; the Smith downloads only what setting it up needs.
With Castellan's developer agents hired, the Smith's page shows more: the model servers by name, new versions to try for your NPU, and checks that the models still answer right. See The Smith, for developers.
Related articles
Is this page right?
If something on it is wrong or out of date, tell us and we'll fix the page.
Still stuck? Write to support@castellan-software.com and mention article 1401. Every version of Smith, and what changed in it, is in its release notes.