Integrations
Manage → Integrations is where administrators choose which registered data source the platform uses for a capability. There are two roles: AI · reasoning, the model used for data quality rules and SQL, questions, search, and FinOps AI insights, and AI · embeddings, the model knowledge bases are embedded with for RAG. This is the only place an AI source or model is set.
Open the role
- Open Manage.
- Select Integrations. This opens the Integrations catalog.
- Select the AI · reasoning or AI · embeddings role row.
The role page has Defaults, Health, and Overrides tabs. The catalog shows the source currently bound to each role and the number of product-specific overrides. Bindings are tenant-wide, so only tenant administrators can save or remove them.
Set the default AI reasoning source
- On the Defaults tab, select the Pick a data source… chip.
- Choose one of the tenant's registered Gen AI Models sources.
- Select Save changes.
The role becomes the global default for products that do not have an override. If the source is not visible to your account, it was registered by another administrator; select a source that you can access before changing the binding.
Add a product override
- Open the role's Overrides tab.
- Select Add override.
- Choose USS or another available product and select its data source.
- Save the override with Add override or Save override.
The product-specific source wins for that product. Select Revert on an override to make the product inherit the global AI reasoning source again. Only tenant administrators can save or remove overrides.
USS and integrations
When the AI · reasoning role is bound, USS uses that role binding. To give search a different model, add a USS product override on the role's Overrides tab.
The USS tab under Manage → Configuration has no AI field. Semantic and hybrid search embed with the AI · reasoning role's USS override, or its global default. Full text search needs no AI source.
Your own setting
AI · embeddings is the one role with a personal layer. Below the tenant-wide Data source card, every user sees a Your own setting card, "Used for the knowledge bases you enable." Binding a source there does not change the tenant default; it only decides what a knowledge base is built with when you are the one enabling RAG on it.
The card shows your own source's name, Tenant default when only the tenant default is bound, or No data source assigned yet. when neither is. Model stays greyed out until you bind a source of your own. Use the tenant default clears your own binding after you confirm.
AI · embeddings has two settings, and you can set both on your own card as well as on the tenant default:
- Model is the embedding model for the bound source. Leave it blank to use the source's own. It belongs to the source it was chosen with, so binding a different source never carries over another source's model. Changing the source or model affects only knowledge bases you enable afterwards.
- Batch size is how many chunks go to the embedding source in one request, from 1 to 1,000, and 50 if you leave it empty. It applies only once a data source is bound; until then the card shows "Your batch size applies once a data source is bound." It makes no difference for a model whose context window is 2,048 tokens or fewer, which receives one chunk per request.
What happens after an upgrade
This screen replaces the older AI settings: the tenant-wide AI Data Source, the separate ones for Data Quality and for search, and the RAG global embedding source and model. Those settings are gone from the product. The upgrade carries what a tenant had configured onto the roles:
- The tenant-wide AI Data Source becomes AI · reasoning's global default.
- Data Quality's own AI Data Source, if set, becomes the Quality override, and search's becomes the USS override.
- The RAG global embedding source, model and batch size become AI · embeddings' tenant default.
A setting is carried over only when the source it names is one of the tenant administrator's own registered Gen AI Models sources. If the source was deleted, belongs to another account, or is not a Gen AI Model, that layer is left unbound rather than guessed; bind it here and the features that need it work again on their next request. A role you had already bound by hand is left alone.
Every knowledge base that already had RAG enabled is fixed to the source and model it was built with. One the upgrade can't match safely stops answering with "This knowledge base couldn't be matched to its embedding source during the upgrade; ask your DataDios administrator to restore it."
When nothing is bound
A feature whose role is not bound does not fall back to anything. It stops and names itself and where to fix it, for example:
- "SQL AI needs an AI source. A tenant admin can set AI · reasoning in Manage > Integrations."
- "Knowledge base RAG needs an embedding source. A tenant admin can set AI · embeddings in Manage > Integrations."
The same applies to AI chat, Insights, Knowledge base chat, Data quality rules and Semantic search. Full text search needs no AI source and keeps working. The AI Studio status pill reads Not Configured, and FinOps AI insights read "AI reasoning isn’t available for this tenant". A role bound to a source that was later deleted stops with "A data source bound to ai.reasoning no longer exists. Rebind it in Manage > Integrations."
Health checks
The Health tab tests the data source bound to a role and reports healthy, drifting, or failed when a test result is available. Select Re-check to run the test again. A source registered by another administrator may not be testable from your account.
Troubleshooting
| What you see | What it means |
|---|---|
| No data source assigned | The role is not bound. Features that need it refuse with a message pointing here until an administrator binds a source. |
| Assigned · not visible to you | The role is bound to a source registered by another administrator. |
| Unavailable | The role settings could not be read or the bound source cannot be used. |
| N overrides | The number of products using a source different from the global role. |
| healthy, drifting, or failed | The latest available health-check result for the role's source. |