This article is auto-synced from its in-app version in Tai.
Company admins only. This page is part of Settings, visible to users with the Company Admin role.
The AI model defaults page shows where your models run, and sets which capability your agents get when they don't name a model themselves.
When you'd use this
Most agents are set to follow the company default rather than pinning a model. This page is where that default lives, so it is the one place to make every such agent faster and cheaper, or slower and more thorough, in a single change. Check the region here too when someone asks where your data is processed.
Steps
In the Workbench sidebar, click Settings.
Under AI & Data, click AI model defaults.
Model region shows where inference runs — Europe or Worldwide. It is read-only: this is a data-residency commitment, so contact your Nezasa representative to change it.
Under Capability, pick the tier agents should use by default — Fast, Balanced, or Most capable. The choice saves as soon as you settle on it. Pick Not set to clear it again, so agents fall back to the standard model.
Tips and limits
Your data stays inside AWS and is never sent to the model providers. Europe keeps inference in our European AWS regions; Worldwide also allows AWS regions outside Europe. The default is Europe, unless you have agreed otherwise with Nezasa.
The default only reaches agents that follow it. An agent pinned to a specific model keeps that model — change it on the agent, under Advanced.
Nezasa's built-in agents always run on the model we ship them with, whatever you set here.
Faster tiers cost less per run. Most capable suits agents doing long, multi-step reasoning; for short lookups it mostly buys you a bigger bill. See Billing Center for what a run costs.
A pinned model can stop being available — a catalogue change, or the region setting moving. When that happens this page lists every affected agent and the replacement model it now runs on. Nothing breaks, but the choice is no longer yours until you pick a supported model on the agent.
Not every model watermarks its text. Some AI providers add an invisible watermark only to text from their newest models. TAI's own labelling doesn't depend on it: reports are labelled AI-generated and carry hidden markers whichever model wrote them.
You can go back to "not set". Picking Not set clears the default again, and agents following it fall back to the standard model.
Related