Skip to main content

The AI Models Page

Choose, review, and keep current the AI models that power your agents, all from one page.

Overview​

Workforce > Intelligence > AI Models replaces the older separate AI Providers and Model Assignments pages. It is where you connect an AI provider, decide which model handles each of your agents' jobs, and keep those models up to date over time.

If you have an old link or bookmark to AI Providers, Model Assignments or a single provider's settings page, it still works. It now opens this page (a link to the rankings tab goes to Workforce > Intelligence > Rankings instead).

New to AI models? Read What Is an AI Model? first. It explains models, providers, levels and jobs in plain language.

What you see depends on who manages your AI providers:

While the page is loading who manages your providers, it shows a spinner. If that check fails, you'll see "We couldn't confirm who manages your AI provider." with a Try again button.

1. Choose a Setup​

The first time you open the page with no setup chosen yet, you see How should your agents think? A setup fills all nine of your agents' jobs with suitable models in one step.

What matters most​

Under the heading are three choices. Picking one does not change anything yet. It explains that route and brings the matching cards forward (the others are dimmed but still work).

  • Keep it inside Microsoft - the models run in your own Azure subscription and are billed on your Azure invoice. A small diagram shows where your agents' AI traffic goes; click it to see it full size, or read Where Your Agents' AI Traffic Goes. This is the choice the page opens on.
  • Select a Frontier Lineup - a ready-made lineup from a leading AI lab, kept current for you. A key explains the four levels a lineup fills: Frontier, Balanced, Fast, and Image.
  • Mix and match - connect several providers and choose the best model for each job yourself.

A short line also explains what a model is, with a link to What Is an AI Model?

If your tenant already has models assigned the old way (before setups existed), a footer note reads "Today you run a hand-built mix". Nothing changes until you choose a setup.

Setup cards​

The cards appear in this order: Microsoft Foundry: GPT, Microsoft Foundry: Claude, then the single providers by name (Anthropic, Google Gemini, Hugging Face, OpenAI, OpenRouter, xAI), then Best of what I have.

Each setup card shows:

  • The provider's logo and name
  • The lineup - which model fills each level today, or "Not included" if the setup has no model for that level
  • List price - a relative tier ($, $$, or $$$), never a dollar amount
  • Average fit - how well the setup's models score, on average, for Reasoning, Composition, and Chat
  • You need - what you still have to do (for example, connect an API key), with a green "Connected" chip once that's done
  • Gap notes in amber, if the setup can't cover every level
  • A Set up [provider name] button

The first card, normally Microsoft Foundry: GPT, carries a Recommended for you ribbon.

Provider cards​

Google Gemini, Hugging Face and OpenRouter have no ready-made lineup, so their cards are for connecting only. Click Connect [provider name], paste the key, and click Check key. After that you choose a model for each job yourself, from Your setup. Models from these providers are not kept current automatically yet.

Once one of these providers is connected, its card shows a "Connected" chip and a Manage [provider name] button for replacing the key or disconnecting.

Not sure?​

If your tenant is on Outermind Credit, a wide Not sure? card appears above the others, offering to keep using Credit instead of picking a provider.

2. The Setup Drawer​

Clicking Set up [provider name] opens a drawer with three steps.

Step 1: Connect​

  • Already connected - a simple "Connected" confirmation, nothing to do.
  • API-key providers - a key field and a Check key button, with "Where do I find this?" instructions specific to that provider. When the key works you see "Key works. Every [provider] model is now in your catalog."
  • Microsoft Foundry setups - if you don't already have a verified Azure AI Foundry connection, a Set up your Azure button starts the Customer Azure Environment setup. Once connected, a checklist shows the model deployments this setup's lineup needs in your region, and whether each is verified yet.
  • Outermind Credit and "Best of what I have" - nothing to connect; this step is skipped.

Step 2: Review Every Job​

A list of the nine jobs your agents rely on, described in plain English and grouped the same way as on Your setup: What your agents do (Thinks and plans, Writes, Chats), Specialty (Creates images), and Behind the scenes (Sorts email, Screens for risk, Summarizes, Reflects overnight, Handles everything else). Each row shows the model the job will use, with its provider's logo, and the model's level and effort on the right.

The first time you set up, when no job has a model yet, that is all you see: the model each job will use. There is nothing to compare with and nothing to keep.

When you already have models, each row also shows today's model, an arrow, the model after you apply, and a price comparison like "1/2 the price" or "same price". You also get:

  • Keep, per row - tick a job's Keep checkbox to keep your current choice for that job; leave it unticked to let the setup take it over. Jobs you've picked yourself before start ticked and read "Keeps your pick"; everything else starts unticked, meaning it's proposed for change.
  • Already on it - a job already running the setup's model for that level shows this instead of a change; applying just makes it follow the setup going forward, without changing the model.

In both cases:

  • Gaps and residency - a job the setup can't cover (for example, Creates images on a provider with no image model) explains why and keeps its current model. A job gated by data residency stays unticked with a residency note until you acknowledge it.
  • Graduation confirmation - if you're on Outermind Credit and apply a provider's setup, you must confirm "This moves you off Outermind Credit for good. Your own <provider> account will be billed." before continuing.

Step 3: Keep It Current​

Choose how new model versions are handled going forward:

  • Always current (marked Recommended) - when a newer model version passes Outermind's qualification and costs no more, you're switched automatically and told about it. You can undo for 30 days.
  • Asks before changes - you're notified when a newer version is ready, but nothing changes until you approve it.
  • Versions managed by you - you manage model versions yourself.

Clicking Apply setup saves your choices and shows progress for each job as it switches over. If a job can't be switched, it's listed with the reason. If someone else changed your setup while you were reviewing, you'll see "Your setup changed while you were reviewing. Check the jobs again." and the review reloads with current data.

3. Your Setup (Steady State)​

Once a setup is applied, the page shows:

  • Summary card - the setup's logo and name and your update policy in plain words. At the top right are two buttons: See what changed and Change setup. Below them, three tiles show how many of the 9 jobs are On setup defaults, your Average quality score, and when it was Last checked.

  • Keep it current - three cards, one per update policy: Always current (marked Recommended), Asks before changes, and Versions managed by you. Each explains what it does. Click a card to switch; the one in use has a heavy border.

  • Jobs, one card per group:

    • What your agents do - Thinks and plans, Writes, Chats
    • Specialty - Creates images
    • Behind the scenes - Sorts email, Screens for risk, Summarizes, Reflects overnight, Handles everything else

    Each row shows an icon for the job, the job in plain English with its technical name underneath, and then, lined up in columns:

    • the model, which you can click to open the model picker
    • a status: Default when the job uses the setup's model, or Your pick: stays current / Your pick: pinned when you chose the model yourself, with a Reset to default link
    • the model's level and, when it has been tested, its fit score for that job

    A model being retired within 30 days shows an amber "Retiring <date>" chip.

  • Connections - one tile per connected provider with its status and last successful check, a Manage button, and an Add a connection button for providers you haven't connected yet.

Managing a connection​

Click Manage on a provider's tile to open a small window with two things you can do:

  • Replace the key - paste a new key and click Check key. The key in use stays until the new one works.
  • Disconnect - your agents stop using that provider, and an agent set to one of its models goes back to your setup's model. You can connect it again any time with a key.

You cannot disconnect a provider while one of your jobs still uses it. The window names the jobs, for example "Writes uses Hugging Face. Choose another model for that job first."

Add a connection asks for a provider's key in the same kind of window. Outermind Credit and Microsoft Foundry are managed on their own pages, so their Manage takes you there, and adding Azure AI Foundry opens its Azure setup page. See LLM Providers for more.

When an update arrives​

If there's a newer model version to tell you about, a banner appears below the summary card, for example: "Moved 4 jobs to Claude Haiku 5: score 79.0 to 85.9, same price. Undo". Click See what changed to open a timeline of every change - upgraded, suggested, now qualifying, needing a deployment, or retiring soon - each with the jobs it affects, the score and price comparison, and actions like Undo, Apply, Skip, Deploy in Azure, or Try on Reasoning.

Changing setups​

Click Change setup to go back to the same page you chose your setup on. Three things are different:

  • A Back to your setup button at the top returns you without changing anything
  • A line tells you which setup you are on today
  • The card for your current setup carries a green Your current setup ribbon, and its button is switched off

Pick another card to open the setup drawer for it, starting at Connect. The Review every job step shows exactly what would change, job by job, with the price comparison. Nothing changes until you click Apply setup.

Your Current Models (Read-Only)​

If your AI providers are managed by Outermind or a partner, or you're in a partner delegated session, you see a read-only view instead: a notice explaining who manages your providers, and the jobs list for all nine jobs (job, model, effort, status, and fit) with no picker, no setup cards, no drawer, no policy control, and no Connections strip. The server refuses any attempt to change these settings while they're managed this way, regardless of what the page shows.

Edge Cases​

  • Load failure - "We couldn't load your AI models." with a Try again button.
  • No providers connected - you see Choose a setup with every setup card showing "Connect" under You need. "Best of what I have" isn't available until you have at least one connected provider.
  • A job with no model assigned - its row reads "Uses Operations: <model>" with no status, until you choose a model for it. Creates images has no fallback, so with no model it reads "Off. Choose an image model."
  • A provider's key stops working - its Connections tile turns amber with "Key not working", and each job on that provider shows a "Provider key not working" chip. Click Manage and replace the key.

FAQ​

I bookmarked the old AI Providers or Model Assignments page. Did I lose my settings? No. The old links open this page, and nothing about your providers or assignments changed.

Where did the page for adding models to a provider go? It was removed. You no longer add models one at a time: connecting a provider makes every approved model it offers available in the model picker. Models you added before still work.

Does choosing a setup change anything I haven't ticked Keep on? No job changes until you click Apply setup, and the Review step shows you exactly what will change before you confirm.

What happens to a job I've customized if I apply a setup? If you ticked Keep for that job, it stays on your chosen model and its status reads Your pick: stays current or Your pick: pinned. Untick it to let the setup take over that job.

Can I go back to picking models individually instead of following a setup? Yes. Open the model picker on any job in Your setup and choose a specific model. Its status switches to Your pick. Use Reset to default on that row to return it to the setup's default.

Can I use a provider that has no setup, like OpenRouter? Yes. Connect it from its card under Mix and match, or straight from the model picker, then choose its models for the jobs you want.