AI providers

Adverity Atlas turns questions into SQL and explanations using a large language model. Settings > Integrations > AI Providers is where you choose which provider and models it uses.

Prerequisites

Admin role at the tenant level. Most workspaces run on the platform-managed default and never need to change anything here.

The managed default

Out of the box, Adverity Atlas uses a provider Adverity manages. The page shows it as Platform-managed default with a Managed badge, naming the provider and model in use and noting that no key is needed and updates apply automatically. Override it only when you have a reason to: your own provider account, a model your policy requires, or data residency rules that mean the model calls must go through your own tenancy.

Note

The managed default card disappears as soon as you configure a provider of your own. From that point the tenant runs its own chain, and the order of that chain is yours to set.

Supported providers

Available Providers lists everything you can configure. Each card shows the provider and the models it covers, and reveals a Configure badge when you hover it.

Provider	Models it covers
OpenAI	GPT-6 Astra, Sol and Luna models
Anthropic	Claude Opus 5.5, Claude Sonnet 5.5, Claude Haiku 5.5 models
Azure OpenAI	Azure-hosted OpenAI models
Google Gemini	Gemini 3.1 Pro, Gemini 3.8 Flash models
Google Vertex AI	Gemini in your own GCP project, keyless IAM auth, region-pinnable
Snowflake (Anthropic)	Anthropic Claude models hosted on Snowflake Cortex
OpenAI-Compatible (Custom)	Any endpoint that speaks the OpenAI API, with API key or OAuth auth

Configuring your own provider

1. Pick a provider: From Available Providers

2. Enter credentials: Stored encrypted

3. Verify: Confirm it works

In detail:

1. Go to Settings > Integrations > AI Providers. Click the card for the provider you want under Available Providers, or Override on the Platform-managed default.

2. Confirm the Provider and give it a Display Name. The name defaults to the provider’s own, so you only need to change it when you are running more than one config for the same provider.

3. Fill in the credential fields for that provider (see below) and save.

4. Open the row’s actions and click Verify. Adverity Atlas calls the provider with its default model in the background and updates Status when the check returns.

Credential fields by provider

Provider	Field
OpenAI	API Key	Required
OpenAI	Organization ID	Optional
Anthropic	API Key	Required
Azure OpenAI	API Key	Required
Azure OpenAI	Endpoint, e.g. https://your-resource.openai.azure.com/	Required
Azure OpenAI	API Version	Optional
Azure OpenAI	Deployment Name	Optional
Google Gemini	API Key	Required
Google Vertex AI	GCP Project ID, the project in your Google Cloud account hosting the model	Required
Google Vertex AI	Location, the Vertex region such as europe-west4, or global / us / eu. Defaults to us-central1, set it explicitly if inference must stay in a particular region	Optional
Google Vertex AI	Service Account Key JSON (not recommended)	Optional
Snowflake (Anthropic)	Programmatic Access Token	Required
Snowflake (Anthropic)	Account Identifier, e.g. xy12345.us-east-1	Required
OpenAI-Compatible (Custom)	Base URL, the OpenAI-compatible base path, usually ending in /v1. Must be HTTPS	Required
OpenAI-Compatible (Custom)	Model, spelled exactly as the endpoint names it. Adverity Atlas passes it through untouched	Required

OpenAI-Compatible (Custom) is the one provider that asks how it should authenticate. Pick an auth mode and only that mode’s fields are shown:

Auth mode	Field
API key	API Key, sent as an Authorization: Bearer header on every request	Required
OAuth 2.0 client credentials	Token URL, your identity provider’s token endpoint. Adverity Atlas exchanges the client credentials here for an access token and refreshes it before it expires	Required
OAuth 2.0 client credentials	Client ID, the application identifier registered with your identity provider	Required
OAuth 2.0 client credentials	Client Secret, the secret paired with the client ID	Required
OAuth 2.0 client credentials	Scope, sent with the token request to say what the token is for	Optional

Use this provider to route Adverity Atlas through a self-hosted model, an internal LLM gateway, or any vendor that exposes the OpenAI API. Because the endpoint names its own model, there is no list to choose from: type the model identifier the endpoint expects.

Tip

On Google Vertex AI, leave Service Account Key JSON blank. The keyless path grants roles/aiplatform.user to the Adverity service account shown after saving, so no permanent credential is stored. Use the JSON field only when you cannot grant a role to an external principal.

Managing the chain

The table lists Provider, Status, and Failover order.

• Primary: The first provider in the list carries this badge. It is the one Adverity Atlas calls first

• Status: Reads Verified, Unverified, or Error. Hover an Error badge for the provider’s own message

• Failover order: The up and down arrows move a provider through the chain, so chat keeps working when one provider is unavailable

• Row actions Verify to re-check credentials, Edit to change them, Delete to remove the provider

Editing a saved provider leaves secret fields blank with “Leave blank to keep existing value”. Tick Remove the stored value to clear one instead of replacing it.

What depends on this

The configured provider powers chat, automations, glossary AI assistance, and artifact generation. A misconfigured provider affects all of them at once, which is why Verify matters before you switch the default.

Note

Changing providers can change how answers read, different models phrase and reason differently. The SQL still runs against your warehouse and the glossary still governs the terms; the wording of explanations is what shifts.

Web research

When someone asks Chat or an automation to research the web, Adverity Atlas uses your AI model’s built-in web search. Settings > Integrations > Web Research is where an admin can configure a dedicated search provider to use instead.

When no search provider is available to connect, the page reads “Using your AI model’s built-in web search” and there is nothing to configure. When one is available, the page shows Add provider:

1. Click Add provider. The Add search provider panel opens.

2. Enter a name, pick the Provider, and enter its API key.

3. Click Add provider. The provider appears in the table with a Status of Verified, Error, or Not verified.

4. Use Verify in the row’s menu to check the key. Edit changes the name, replaces the API key (“Leave blank to keep the existing key.”), or clears Active to disable the provider. Delete removes it.

Anyone who can open Settings sees the page; adding, editing, verifying, and deleting providers needs the Admin role. Who can run web research is a separate permission held by Editors and above, see Chat.

Permissions

Tenant Admin only. Credentials are never shown again after saving.

What’s next?

• Chat, what the provider powers day to day

• Cost monitoring, track spend per provider and model

• Service configs, the other Integrations settings

Was this page helpful? Yes No