How to connect an AI model provider to DataLens
A model provider is a connection like any other — credentials, an environment, a status. You supply the key, you choose which models are allowed.
What you will do
You will add an AI model provider using your organisation's own credentials, restrict the connection to a specific set of allowed models, choose a default, and confirm it works with a real call.
AI features call the provider whose credentials you supplied. Restricting the allowed model list is how you control what those features can reach for.
When this is useful
- You want to use the AI-assisted features and your organisation requires its own provider account.
- You need to restrict which models can be used, for cost or policy reasons.
- You want AI usage attributable to an account you control and can revoke.
Before you start
- A provider account and API credentials — Issued by the model provider, not by DataLens.
- A decision on which models to allow — The connection can be restricted to a specific list rather than everything the provider offers.
- Awareness that testing costs money — The test makes a real call to the provider. The interface says so explicitly.
Steps
Open Data → Connections → AI Model
Select Data, then Connections, then the AI Model tab.
A model provider sits here rather than on a screen of its own because it is a connection: it has credentials, an environment and a status like every tab beside it.
Enter the provider credentials
Give the connection a name and supply the credentials your provider issued.
Set the Environment — production or otherwise — so the connection can be told apart from a test one later.
Load and choose the allowed models
Load the model list from the provider, then tick the models this connection is permitted to use. The result is the Allowed models list.
Restricting the list is the control point: a model not on it cannot be used through this connection.
The chosen models listed as allowed.
Set a default model
Choose the Default model — the one used when a feature does not name a specific model.
The default must be one of the allowed models. Unticking a model that is currently the default clears the default.
Test with a real call
Test the connection. This makes a genuine call to the provider and therefore spends money — the interface warns you before you do it.
A successful test proves the credentials and the model choice together, which is the only thing that actually proves the connection works.
A confirmation that the provider answered, or a specific failure.
Save the connection
Save it. AI-assisted features can now use it, subject to the allowed model list you set.
What happens next
AI-assisted features across the platform can now use this provider. What they can reach for is bounded by the allowed model list.
Usage is observable: Govern → AI Operations reports cost, tokens, failures, cache behaviour and estimate versus actual for the AI pipeline.
Example
A data platform team adds their own provider key, allows two models — a small one for routine work and a larger one for harder tasks — and sets the small one as default. Cost stays predictable, and AI Operations shows them where the spend actually went.
Tips
- Restrict the allowed models deliberately. It is the simplest cost control available and it is set once.
- Set the default to your cheaper model. Features that do not name a model will use it.
- Remember the test spends money. Test once when you set up, not repeatedly.
- Check Govern → AI Operations after a period of use. Estimate versus actual is the number that tells you whether your assumptions held.
Limitations
- You supply the credentials. AI calls go to the provider whose key you configured, on your account.
- The test makes a real call to the provider and therefore costs money.
- A default model must be one of the allowed models — removing it from the allowed list clears the default.
- The available providers are those offered on the AI Model tab.
Related how-to guides
Build a document collection
Load Word, Markdown, HTML or text documents as a collection — stored whole, not parsed into rows.
Run a scenario simulation
Ask what happens if a variable changes — forecasts, shocks, comparisons and impact, on your own data.
Generate a data document
Produce an architecture document, a model spec, a mapping or a runbook from what you have already built.
Related questions
More of these on the DataLens FAQ page.
Does DataLens use its own AI models or mine?
AI calls go to the provider whose credentials you configured on the AI Model connection tab. You supply the key, you choose which models the connection is allowed to use, and usage is on your provider account.
Can I limit which AI models DataLens is allowed to use?
Yes. The connection holds an allowed models list that you tick explicitly, plus a default model used when a feature does not name one. A model that is not on the allowed list cannot be used through that connection.
Where can I see what AI usage is costing?
Govern → AI Operations reports on the AI pipeline: cost, tokens, latency, failure taxonomy, quarantine, cache behaviour and estimate versus actual.
Try this in DataLens
DataLens is in private beta. Request access and work through this guide on your own data.
Request beta access