Provider Setup

This page walks through the specific configuration for each supported AI provider. Each section shows the provider block you would add to your llm-config.json file and the environment variable needed for authentication.

For an overview of the configuration file format and how API keys work, see LLM Configuration.

OpenAI

The most common setup. Use this if you have an OpenAI API account.

{
  "ProviderId": "openai",
  "BaseUrl": "https://api.openai.com/v1",
  "ApiKeySecretName": "OPENAI_API_KEY",
  "IsLocal": false,
  "Models": [
    {
      "ModelId": "gpt-4o",
      "MaxInputTokens": 128000,
      "MaxOutputTokens": 16384,
      "Capabilities": "Chat,Tools"
    },
    {
      "ModelId": "gpt-4o-mini",
      "MaxInputTokens": 128000,
      "MaxOutputTokens": 16384,
      "Capabilities": "Chat,Tools"
    }
  ]
}

OPENAI_API_KEY=sk-your-key-here

If your organization uses an OpenAI organization ID, add it to metadata:

"Metadata": {
  "Organization": "org-your-org-id"
}

Google Gemini

Gemini uses Google's OpenAI-compatible endpoint, which means the same protocol as OpenAI. The key differences are the base URL and a metadata entry for the chat completions path.

{
  "ProviderId": "gemini",
  "BaseUrl": "https://generativelanguage.googleapis.com/v1beta/openai",
  "ApiKeySecretName": "GEMINI_API_KEY",
  "IsLocal": false,
  "Metadata": {
    "ChatCompletionsPath": "/chat/completions"
  },
  "Models": [
    {
      "ModelId": "gemini-2.5-pro",
      "MaxInputTokens": 1048576,
      "MaxOutputTokens": 65536,
      "Capabilities": "Chat,Tools"
    },
    {
      "ModelId": "gemini-2.5-flash",
      "MaxInputTokens": 1048576,
      "MaxOutputTokens": 65536,
      "Capabilities": "Chat,Tools"
    }
  ]
}

GEMINI_API_KEY=your-gemini-key-here

Large Context Window

Gemini 2.5 Pro supports a 1M token input window — significantly larger than other providers. This makes it particularly well-suited for workflows that process large documents or long conversation histories.

Anthropic (Claude)

Anthropic uses its own Messages API format. The platform handles the format differences automatically — you configure it the same way as any other provider, and the Anthropic client translates system messages, tool calls, and response parsing behind the scenes.

{
  "ProviderId": "anthropic",
  "BaseUrl": "https://api.anthropic.com",
  "ApiKeySecretName": "ANTHROPIC_API_KEY",
  "IsLocal": false,
  "Models": [
    {
      "ModelId": "claude-sonnet-4-20250514",
      "MaxInputTokens": 200000,
      "MaxOutputTokens": 8192,
      "Capabilities": "Chat,Tools"
    }
  ]
}

ANTHROPIC_API_KEY=sk-ant-your-key-here

Azure OpenAI

Azure OpenAI requires your resource URL and deployment details. Authentication uses the api-key header instead of Bearer tokens.

{
  "ProviderId": "azure-openai",
  "BaseUrl": "https://your-resource.openai.azure.com",
  "ApiKeySecretName": "AZURE_OPENAI_API_KEY",
  "IsLocal": false,
  "Metadata": {
    "ApiVersion": "2024-02-15-preview",
    "DeploymentId": "your-deployment-name"
  },
  "Models": [
    {
      "ModelId": "gpt-4o-deployment",
      "MaxInputTokens": 128000,
      "MaxOutputTokens": 16384,
      "Capabilities": "Chat,Tools"
    }
  ]
}

AZURE_OPENAI_API_KEY=your-azure-key-here

The platform automatically builds the correct deployment URL path from the metadata values. If you need a custom path, use ChatCompletionsPath in the Metadata instead of DeploymentId.

Enterprise Compliance

Azure OpenAI is the right choice if your organization requires data to stay within your Azure tenant or you have specific compliance requirements. The models run in your own Azure subscription.

Ollama (Air-Gapped / Self-Hosted)

For air-gapped deployments where no external API calls are permitted, you can run models locally using Ollama or any OpenAI-compatible local server. Set IsLocal to true so routing strategies can distinguish local providers from cloud providers.

{
  "ProviderId": "ollama",
  "BaseUrl": "http://ollama:11434",
  "ApiKeySecretName": "OLLAMA_API_KEY",
  "IsLocal": true,
  "Models": [
    {
      "ModelId": "llama3.1:70b",
      "MaxInputTokens": 128000,
      "MaxOutputTokens": 4096,
      "Capabilities": "Chat"
    }
  ]
}

Air-Gapped Deployments

Local models typically do not support tool calling. If you need agent features (MCP Server, autonomous agent nodes), you will need at least one cloud provider with a model that has the Tools capability, or a local model server that implements the OpenAI function calling protocol.

Using Multiple Providers

You can configure any number of providers simultaneously. A typical multi-provider setup:

Google Gemini 2.5 Pro for large-context analysis (1M token window)
OpenAI GPT-4o for agent tasks (reliable tool calling)
OpenAI GPT-4o Mini for high-volume, low-cost classification
Ollama as a local fallback for air-gapped environments

Each profile independently selects which provider and model to use. See Profiles & Routing for how to set this up.

Next Steps

LLM Configuration — overview of the config file format and API keys
Profiles & Routing — balance cost and quality, configure per-node overrides
AI-Powered Workflows — add AI steps to your workflows