AI providers

Admin → AI providers is where you manage the default AI connections your whole organization shares (Organization AI Defaults). Personal account connections are managed in each person's own Settings → My AI Connections.

Each connection is one way of reaching a model — a signed-in account, an API key, or a self-hosted server.

Adding a connection

Add connection offers two methods:

• OAuth — sign in with an existing ChatGPT or Claude account. No key to manage. (Use an API key for Google models.)
• API key — paste a key and choose a model. The provider dropdown is grouped into Cloud (OpenRouter, OpenAI, Anthropic, Google) and Local / self-hosted (Ollama, LM Studio, vLLM).

For OAuth, press Add account on the ChatGPT or Claude card and finish signing in with Open — the connection is saved automatically (if you're asked for an authorization code, enter it and press Submit code). For API key, press Save connection once you've filled in the form.

Connecting a self-hosted model

Pick a local provider (vLLM, Ollama, or LM Studio) in the API key form and three more fields appear:

• Base URL — where your model server lives (for example http://host.docker.internal:11434).
• Context length (optional) — for vLLM, your --max-model-len value.
• Max output tokens (optional).

Then press Check. Ian On probes the server, lists the models it found, and selects the first one for you. If the model you want isn't in the list, the Enter manually… option is always there as a fallback.

With Validate on (it's on by default), pressing Save connection checks that the server actually serves that model and whether the model supports image input. A model without image support is flagged in the save notice and with No image support in the table. A local model is used for tasks that need images only when its image support has been confirmed.

Connections and use order

Your connected AI models are collected in the Organization defaults table — Provider / account · Connection type · Model · Use order · Status · Usage · Actions. Opening the screen checks each connection's status and usage, and Refresh status checks again. The Status column tells apart states such as Reconnect required, Quota exhausted, and Rate limited. Live check unsupported (API key connections) and Status unavailable don't mean the connection has failed.

For an OAuth connection, you can switch to another model that account offers from the table's Model column. The list opens once the status check finishes after you open the screen. If the list can't be loaded, the saved model stays as it is, and the Model column may show Couldn't load models. A connection whose status is Reconnect required needs you to sign in again before its list opens. API key and local connections show only the model name.

• Use order — set the order with the up and down arrows (Move up / Move down). The Rank 1 connection is used first and is marked Always use. From rank 2 on, only connections with Fallback turned on are switched to automatically. If a member has added personal connections in My AI Connections, that member's work uses all of them first, then this order.
• Automatic switching — if expired authentication, a usage or request limit, or a connection outage hits before the response starts, Ian On moves on to the next Fallback connection. Errors after the response has started, and any other errors, don't switch; if there's no connection left to switch to, the request ends in an error. A request that names a specific model only switches between connections set up with the same provider and model.

Voice transcription

Lower on the same screen is a separate Voice Transcription card. Here you Connect the organization-wide API key that Meetings uses for live dictation. Voice transcription is separate from the AI model connections and use order used for answers. If it isn't connected, meetings run on demo transcription.

Common situations

• "The model just stopped responding." On this screen, press Refresh status to check status and usage. If it shows Reconnect required, sign in to the same account again with Add account for an OAuth connection, or save an API key connection again with a new key. For when automatic switching happens, see Use order above.
• Meeting live dictation won't turn on. Check that the Voice Transcription card below shows Connected · live transcription on. If it shows Not connected · demo transcription, meetings run on demo transcription. If Meetings isn't in the left menu, turn on Meetings in Admin → Features.
• Token limits aren't set here. The Usage column on this screen shows the account usage your provider reports, or, for a self-hosted connection, the server's current capacity. Per-request and daily token limits are set per agent on the Agents screen.