Configure a project's models and AI settings#
Choose which models a project can use and set its standing instruction and default settings.
Before you begin#
- To change the model list, you need the Owner, Admin or Account Manager role. A project manager can't.
- To change the configuration, you need one of those roles or the project manager role.
- The project is active.
Choose the models#
- Open Projects, select the project, and open the Models tab.
- Select Add models, then pick from 1 to 100 models from the catalogue. The picker hides models already on the list.
- Select Add N models, where N is the number you picked. Adding the same model twice does no harm.
To remove models, tick them and select Remove N, where N is the number you ticked. ATLAS asks for no extra confirmation.
Result: the project can reach only the models on its list. With an empty list, the gateway answers that the project has no models. A model deactivated in the catalogue stays on the list as unavailable. Requests for it fail with "model not found" until it's reactivated.
Set the standing instruction and defaults#
- Open the Configuration tab.
- Fill in Standing Instruction, for example "You are a helpful assistant. Answer in English only.", plus Default Temperature and Default Max Output Tokens. Empty fields read "No default set".
- Save.
Saving replaces the whole configuration, so any field you leave empty is cleared.
Good to know#
- The standing instruction is added to requests made with the project's own API keys. It isn't added to requests made with personal keys acting under the project.
- Defaults apply only when the caller leaves the value out.