Adding models

A model that hasn’t been added can’t be used anywhere in your organization. The Models page lists only the models you have added, and Add model opens the catalogue of everything else Intric offers. Adding a model is what makes it available to your colleagues.

Why adding models matters

Models start switched off. A new organization starts with a small starter set, roughly one model of each type. Every other model in the catalogue stays off until an Admin adds it. If people report that a model they’ve heard of isn’t in their model picker, the usual reason is that nobody has added it yet.

Adding a model makes it available across the organization. An added model appears in every Space whose security classification it meets, with no separate step for each Space. Removing it takes it away from all of those Spaces at once.

It decides what the curated picks can choose from. Balanced, Fast, Economical and Smart only ever pick from the models you have added. With only a few models added, all four may resolve to the same model. With a wider selection, each pick can find a model that suits its priority. Because a pick is worked out again each time an assistant is used, adding or removing a model can change which model an assistant set to a pick runs on.

It is your control over where data goes. Each model sends data to a sub-processor and runs in a particular hosting region. Deciding which models to add is how you keep usage to the vendors and regions your organization has approved.

Adding a model

Add a model from the catalogue

1. Go to Admin → Platform → Models and select Add model.
2. Choose a filter in the sidebar to see models. The list stays empty until you pick one.
3. Select the + (or Add) on a model. It changes to Added, and the model appears on the Models page straight away.
4. Add as many models as you need, then close the dialog. To take a model out again without leaving the dialog, select Remove on a model you added.

Finding the right model

The sidebar narrows the catalogue by:

• Type — completion, embedding, transcription or image.
• Intelligence and Speed — a minimum benchmark score out of 100.
• Cost — a maximum cost tier.
• Capabilities — vision, reasoning, and tool use.
• Hosting region — where the model runs.
• Sub-processors — who receives the data when the model runs.
• Company — the vendor behind the model.

Results are sorted by intelligence, with the most capable models first. You can switch between cards and a table, and search the list by name, vendor or type. Select the info icon on a model to open its model card before you decide.

Note

Only completion models have intelligence, speed and cost ratings. Raising the Intelligence or Speed slider, or capping Cost, therefore hides every embedding, transcription and image model. Leave those three filters at Any when you’re looking for one of those types.

Models you can’t add

A dimmed model with no add button can’t be enabled in your organization yet. Hover over it to see why. Usually its sub-processor hasn’t been approved: enable the sub-processor under Sub-processors, and the model can then be added. Tick Hide unavailable to leave these models out of the list.

After adding a model

Each model on the Models page has an actions menu:

• Set as default — makes it the model new assistants start with, one default per model type. The personal assistant also falls back to it when users aren’t allowed to choose their own model.
• Edit security classification — decides which Spaces can use the model. This appears when security classifications are turned on. See Model classification.
• Remove — takes the model out of your organization again.

Removing a model

When you remove a completion model that assistants or widgets use, Intric asks what should take over before it removes the model. The dialog shows how many assistants and widgets use the model and in how many Spaces. Under Replace with, choose a curated pick such as Balanced, or switch to Specific model and choose one. Replace and remove moves every one of those assistants and widgets to the replacement and then removes the model, so none of them stops answering.

The dialog only offers replacements that meet the security classification of every affected Space. If an affected assistant or widget uses tools, it only offers models that support tool use. A widget can’t be set to a curated pick, so it moves to the model that pick chooses in the widget’s own Space. If the model you remove is the organization’s default, the replacement becomes the default.

Personal assistants move too. If any personal assistant uses the model, or it is the default model for personal assistants, the dialog asks for a replacement and says it takes over for them. The replacement has to meet the security classification set for the default personal assistant, and support tool use if personal assistants have tools or image generation. Every personal assistant keeps the model it was created with, even while users aren’t allowed to choose their own, so the dialog can ask for a replacement when model selection is turned off.

Assistants and personal assistants show the new model straight away, without reloading the page.

The same step opens when you select Remove on an added completion model in the Add model dialog. If nothing uses the model, the dialog says so and Remove takes it out.

A removed Intric model isn’t deleted. It moves back to the catalogue, and you can add it again later.

Removing a custom model deletes it permanently, and Intric won’t delete one that is still in use. See Custom models.

Adding a model Intric doesn’t offer

If your organization runs its own model, or has a private deployment with a provider, you can connect it as a custom model with Add custom in the same dialog. See Custom models.