Models

The Models page lists every model generation, Explore and chat can use (#441). Open it from Models in the main navigation. It is there for anyone who may generate in some workspace, and in demo mode, where it lists the server’s models only.

The page has three sections.

Server

The models the operator declared, in AI_MODELS_FILE or the AI_* variables. Each row has its name, provider, model id and where its credentials come from, such as “From ANTHROPIC_API_KEY”. A model whose variable is not set says which one. The server’s default is marked. The section is read-only: an operator changes it by editing the file and restarting.

Each workspace

One section per workspace you generate in, with its own models and which one a generation there uses when nobody picks.

A source-admin of the workspace can:

• Add model: give it a name, choose a provider, and fill in what that provider asks for. See the provider kinds.
• List models, in the dialog once the key is in: ask the provider which models the key may use. Clicking or typing in the Model field then shows them, narrowed by what you type (an id or a display name); pick one, or keep typing any id. Azure deployments and Bedrock models are typed by hand.
• Test: make one small model call, so a wrong model id or key is found now rather than on someone’s next generation. The call counts against the workspace’s limits.
• Edit a model’s name or settings.
• Set as default, or Use the server’s default.
• Delete a model, after confirming. Deleting the default goes back to the server’s.
• Turn Let people use their own models here on or off.

Everyone else sees the workspace’s models and its default, and cannot change them.

Yours

Your own models, billed to you, with Add model, Edit, Test and Delete. They can be used only in workspaces whose source-admins allow personal models; the section says which those are. Elsewhere they are kept and ignored.

Picking a model

Every place that asks the model something has a model chip beside the source chip (#442): a new dashboard and its follow-ups, the panel editor’s “Describe a change”, the source drafter on Data sources, Explore, and the dashboard chat. It shows the model the next request will use. Its menu lists the models you may use in that workspace under Server, Workspace and Yours; one that cannot be called is greyed out with the reason. Manage models… opens this page.

• Your pick is remembered per workspace, on every device: the next new dashboard or conversation there starts on the model you used last.
• A conversation keeps its model. An Explore conversation, or a dashboard’s chat, uses the model it was asked with for every later turn and after a reload, until you change the chip. If that model is deleted, the next turn uses your default instead.
• Without a pick, a request uses your last pick in that workspace if you can still use it, else your first personal model where personal models are allowed, else the workspace’s default, else the server’s.

Each generation’s audit row names the model it used.

Credentials

A key is encrypted at rest and never shown again. The page says only whether one is stored and its last four characters. Leaving a key field blank when editing keeps the stored key, and “Remove the stored key” clears it. Changing where requests go, such as another provider, another base URL host, an Azure resource or an AWS region, asks for the key again. The full rules are in Credentials.

Settings → Workspace model and Settings → Personal model, from before this page, now open it.