AI Provider Setup

Ragtime supports multiple AI providers for powering your assistants. This guide walks you through configuring each provider, with detailed instructions for Azure OpenAI.

Supported Providers

Provider	Description	Setup Complexity
OpenAI	Direct access to OpenAI models (GPT-4.1, o-series, etc.)	Simple — just an API key
Azure OpenAI	OpenAI models hosted on your own Azure infrastructure with enterprise compliance	Moderate — requires Azure resource + model deployments
Mistral	Mistral AI models	Simple — API key from Mistral console
AWS Bedrock	Models via Amazon Bedrock	Moderate — requires AWS credentials

You configure your AI provider in Dashboard → Organization → AI Settings.

OpenAI Setup

1. Go to platform.openai.com/api-keys and create an API key.
2. In Ragtime, navigate to Dashboard → Organization → AI Settings.
3. Select OpenAI as the provider.
4. Paste your API key and click Save.

That’s it — model dropdowns will let you pick from available OpenAI models directly.

Azure OpenAI Setup

Azure OpenAI gives you the same OpenAI models but hosted on Microsoft Azure, which can be important for data residency, compliance, or enterprise requirements. The setup requires a few more steps than direct OpenAI.

Step 1 — Create an Azure AI Services Resource

1. Sign in to the Azure Portal.
2. Click Create a resource and search for “Azure AI services” (previously called “Azure OpenAI”).
3. Click Create and fill in:
   • Subscription: your Azure subscription.
   • Resource Group: pick an existing group or create a new one.
   • Region: choose a region close to your users (e.g. West Europe, East US). Note: not all models are available in every region.
   • Name: give it a recognisable name (e.g. ragtime-ai-prod).
   • Pricing Tier: select Standard S0.
4. Click Review + Create, then Create. Wait for the deployment to complete.

Step 2 — Copy Your Endpoint and API Key

1. Once the resource is created, go to its overview page.
2. In the left sidebar, click Keys and Endpoint (under Resource Management).
3. Copy KEY 1 (or KEY 2 — both work) and the Endpoint URL.
4. The endpoint looks like: https://your-resource-name.openai.azure.com

Important: Copy the resource-level endpoint (shown on the Keys and Endpoint page), not a project-level endpoint from AI Foundry. The resource endpoint is the one Ragtime needs.

Step 3 — Deploy Models in Azure AI Foundry

Azure OpenAI doesn’t let you call models by name like the OpenAI API does. You must first deploy each model, which gives it a deployment name you choose.

1. Go to Azure AI Foundry (ai.azure.com) and select your resource.
2. In the left sidebar, go to Build → Deployments (or My Assets → Models + Endpoints).
3. Click + Deploy model → Deploy base model.
4. Select the model you want (e.g. text-embedding-3-small) and choose Standard deployment type.
5. Give it a deployment name. Tip: use the same name as the model for clarity (e.g. text-embedding-3-small).
6. Set the Tokens per Minute (TPM) rate limit based on your needs, then click Deploy.

You need to deploy at minimum:

Purpose	Recommended Model	Example Deployment Name
Embedding (for knowledge base)	text-embedding-3-small	text-embedding-3-small
Chat / main LLM	gpt-4.1-mini or gpt-4.1	gpt-4-1-mini

Optionally, you can also deploy separate models for:

• Query rewriting — a fast model for reformulating user queries before RAG search (e.g. gpt-4.1-nano).
• Vision — a vision-capable model for processing images (e.g. gpt-4.1).

Step 4 — Handle Model Availability and Waitlists

Not all models are available immediately in Azure. Some newer models (GPT-5 series, o-series reasoning models) require you to register or join a waitlist before they become available for deployment.

• If a model shows a “Register” button instead of “Deploy”, click it and fill in the registration form. Approval can take days or weeks.
• Model availability varies by region. If a model isn’t listed, try a different region or check the Azure OpenAI model availability table.
• Stick with generally available models like gpt-4.1-mini, gpt-4.1, and text-embedding-3-small to avoid waitlist delays.

Step 5 — Configure in Ragtime

1. In Ragtime, go to Dashboard → Organization → AI Settings.
2. Select Azure OpenAI as the provider.
3. Paste your API Key (from Step 2).
4. Paste your Endpoint URL (from Step 2).
5. For each model field (Embedding, Chat, Query Rewrite, Vision), type the exact deployment name you chose in Azure AI Foundry.
6. Click Save, then use the Test Connection button to verify everything works.

Common Pitfalls

Problem	Cause	Solution
404 — Resource not found	Wrong endpoint URL or model not deployed	Double-check the endpoint from Keys and Endpoint. Verify the deployment exists in AI Foundry.
401 — Unauthorized	Invalid or expired API key	Re-copy KEY 1 from the Azure portal and update it in Ragtime.
400 — Unavailable model	Using a model name instead of a deployment name	Azure requires deployment names, not model identifiers. Enter the exact name you gave the deployment in AI Foundry.
Knowledge base ingestion fails	Embedding deployment not found or provider mismatch	Ensure your embedding deployment name matches exactly. Note: libraries lock their embedding config at creation — if you switch providers, create a new library.
Model not listed in AI Foundry	Region doesn’t support it, or it requires registration	Check availability by region. Try East US or Sweden Central which tend to have the widest selection.

Azure OpenAI for Voice / Realtime

If you use voice-enabled assistants, you can also configure Azure OpenAI for the realtime voice backend separately under Dashboard → Organization → AI Settings → Voice. The same principles apply:

1. Deploy a realtime-capable model in AI Foundry (e.g. gpt-4o-realtime-preview).
2. Use the same endpoint and API key (or a separate resource if preferred).
3. Enter the deployment name in the Voice settings.