Skip to main content
RAUM uses your own AI provider account. During onboarding, you connect provider credentials and select the models RAUM should use for generation and semantic retrieval.

​
Supported providers

The onboarding wizard currently includes:
  • OpenAI
  • Anthropic
  • Gemini
  • Azure OpenAI
  • DeepSeek
  • Groq
  • Mistral
  • AWS Bedrock
  • Vertex AI
  • Moonshot / Kimi
  • Qwen
  • OpenRouter
  • Custom OpenAI-compatible endpoints
The model list is fetched from the provider or RAUM’s current provider catalog. Model names change frequently, so this documentation does not maintain a static list of recommended model IDs.

​
What you configure

The AI Engine step can collect:
  • Provider
  • API key or provider credential
  • Base URL, when the provider or custom endpoint requires one
  • All-purpose/chat model
  • Embedding model
  • an optional secondary provider when the primary provider cannot supply every required model purpose
RAUM validates tenant-controlled provider URLs before storing or calling them. Custom base URLs must use an allowed public HTTPS destination; local, loopback, metadata-service, and other private-network destinations are rejected during public onboarding.

​
Connect the provider

1

Choose your provider

Select the provider you want RAUM to use. The wizard adjusts the available fields for the selected provider.
2

Enter the credentials

Paste the API key or other credential required by the provider.For providers that need a custom endpoint, enter the Base URL shown by your provider.
3

Verify and fetch models

Use the verification control in the wizard. RAUM checks the provider connection and retrieves or classifies the models available to the account.
4

Choose the all-purpose model

Select the model RAUM should use for support reasoning and response generation.
5

Choose the embedding model

Select the embedding model used for semantic retrieval over procedures, documents, websites, and other searchable knowledge.
6

Add a secondary provider if required

If the primary provider does not expose a suitable embedding model, or otherwise cannot satisfy a required model purpose, the wizard can ask for a second provider.Verify that provider as well and choose the missing model from its available list.

​
Why RAUM needs two model purposes

​
All-purpose / chat

This model handles the generative parts of the support workflow, including understanding requests, reasoning through procedures, and producing responses.

​
Embeddings

Embeddings convert knowledge into vectors that RAUM can search semantically. They are used to retrieve relevant procedures and knowledge before a generative model writes an answer. These are different workloads. A provider can be excellent for generation while offering no embedding model at all, which is why RAUM supports a secondary provider.

​
Credential handling

Provider credentials are stored encrypted by RAUM and are not returned back to the browser as plaintext after they are saved. RAUM uses those credentials when it needs to call the provider for your organization’s workloads.
Bring-your-own-provider does not mean support data bypasses RAUM’s backend. RAUM coordinates the request, applies your organization configuration and knowledge, and sends the necessary model request to the provider you selected.

​
Built-in versus custom base URLs

For built-in providers, RAUM can apply the provider’s canonical API endpoint when appropriate. Use a custom base URL only when your provider configuration actually requires one, such as:
  • an Azure or cloud-specific deployment endpoint;
  • an API gateway you control;
  • OpenRouter or another compatible routing service; or
  • a custom OpenAI-compatible provider.
Do not enter localhost, private IP addresses, cloud metadata endpoints, or internal-only hostnames during public onboarding.

​
After onboarding

Administrators can review or change model configuration from the AI workspace in the RAUM dashboard. Changing a model affects subsequent AI requests. If you rotate a provider key, update it in RAUM before revoking the old credential if you want to avoid an interruption.

Create and verify your account

See how provider setup fits into staged registration and email verification.

Finish Chatbot setup

Confirm an active model before creating your first web channel.