Skip to main content
Help Center

Connect an AI

Enter the tenant's AI endpoint: endpoint, model, access key and maximum input length – after that the AI features are ready to use.

Prerequisites

  • Administration rights for the system configuration
  • The endpoint address and access key of an AI provider – or of a self-hosted model (see Your AI, your data)

teamspace brings no AI of its own. Before any AI feature appears, an administrator has to enter the endpoint that teamspace sends its requests to, once. That is the only setup – after that it is only a question of which features are enabled.

Why it is done this way and what you should check when choosing a provider is explained in Your AI, your data. This article shows the mechanics.

Where the setting lives

Configuration → Interfaces → Actions → AI integration

The AI integration is not a tab of its own but an action in the left-hand area of the Interfaces category – below the category list in the Actions section.

The Interfaces configuration category with the list of set-up interfaces; at the bottom left in the Actions area, the AI integration entry is highlighted. The Connection column is redacted because it contains access data.
The "Interfaces" category with the Actions area at the bottom left and the "AI integration" entry

Enter the AI

  1. Open Configuration → Interfaces and click AI integration on the left under Actions.

  2. Fill in the fields:

    FieldRequiredMeaning
    Enabled–The main switch. Without it, teamspace calls no AI – whatever the other fields contain.
    EndpointyesThe address teamspace sends its requests to. This can be a provider on the internet or a model in your own data centre.
    ModelyesThe model identifier your provider expects – written exactly as it documents it.
    API key–Your provider’s access key. For a self-hosted model without authentication, the field stays empty.
    Maximum input length (characters)yesHow many characters a single call may transmit at most.

  3. Save.

The AI integration dialog above the Interfaces category: the Enabled switch is on, below it the Endpoint, Model, the empty API key and Maximum input length (characters) fields with the value 50000, at the bottom the Save and Cancel buttons; at the bottom left the Actions area with the AI integration entry is visible
The "AI integration" dialog with the Enabled switch, Endpoint, Model, API key and maximum input length

Understanding the maximum input length

This field is easily overlooked, even though it governs two things at once.

It caps the cost. Almost all providers bill by the volume transmitted. A ticket with four hundred entries costs many times more than one with four. The limit prevents a single outlier from shaping the monthly bill.

It caps what goes out. Every character above the limit is not transmitted. That is a data protection setting too, not just a cost brake.

A common starting value is in the range of a few tens of thousands of characters – around 50,000. That covers long ticket histories and extensive reports. Don’t set it too tight: if input is truncated, the AI works from an incomplete picture and delivers a summary that gives no sign that half of it was missing.

Check the value against your model. Every model has its own upper limit for input. If your setting is above it, the provider rejects the call and the feature fails instead of trimming.

Check that it works

After saving, the enabled AI features appear in their places. The quickest test:

  1. Open a ticket with a longer history.
  2. Call up the summary – see Condense a ticket history with AI.
  3. If a sensible summary comes back, the endpoint, model and key are set correctly.

Common questions & needs

You want to …How to
Switch AI off completelyTurn off the Enabled switch. Takes effect immediately for all features and all users.
Use a different modelChange the Model field and save. No further action needed.
Ensure no data leaves your premisesRun a model in your own data centre and enter its address as the Endpoint.
Cap the costLower the maximum input length. In addition, limit the allowance at the provider itself.
Know why a feature doesn’t appearCheck Enabled first, then the endpoint and model, then whether the feature itself is enabled.
Use a model without authenticationLeave the API key empty – for a self-hosted model that is the normal case.

Notes

  • One endpoint applies to the whole tenant. Not per user, not per module, not per feature.
  • The key is yours. teamspace uses it for the calls; allowance and billing run through your contract with the provider.
  • No connection, no AI features. They don’t appear at all – to users it looks as if they don’t exist.
  • This applies only to the built-in features. If you connect your own assistant via MCP, you use that assistant’s AI and not this setting at all.