AI Providers
AI providers connect Raikoo to external LLM services, making models available to agents, workflows, and chat sessions. This guide walks through creating and configuring providers, managing their models, and setting up model families.
Understanding AI Providers
AI providers are organization-level connections to external AI services. Each provider holds a type, API credentials, and an optional base URL. Once configured, the provider's models become available for selection throughout the organization.
For a conceptual overview of how providers, models, and model families relate to each other, see Models.
Permissions Required
You need the ai-providers organization permission to manage AI providers. See Roles & Permissions for details.
Accessing AI Providers
- Navigate to your organization dashboard
- Click AI Providers in the left sidebar under the Configuration section
- You will see a list of all configured AI providers for your organization
Adding a New AI Provider
- From the AI Providers list, click Create
- Complete the fields on the Configuration tab:
| Field | Required | Description |
|---|---|---|
| Provider Type | Yes | The AI service this provider connects to (e.g., OpenAI, Anthropic) |
| Name | Yes | A descriptive label for this provider (e.g., "OpenAI Production") |
| Description | Yes | Brief description of this provider's purpose |
| API Key | No | Authentication key for the provider's API. When editing an existing provider, leave empty to keep the current key |
| Base URL | No | Custom endpoint URL. Required for Azure OpenAI, Ollama, and OpenAI-Compatible providers |
- Click Save to create the provider
- After saving, the Models tab becomes available
Multiple Providers
You can add multiple providers of the same type. For example, separate OpenAI providers for different accounts, teams, or billing tiers.
Supported Provider Types
| Provider Type | Notes |
|---|---|
| OpenAI | Connects to OpenAI's API. Requires an API key |
| Anthropic | Connects to Anthropic's Claude models. Requires an API key |
| Google Generative AI | Connects to Google's Gemini models. Requires an API key |
| Azure OpenAI | Connects to Azure-hosted OpenAI deployments. Requires a Base URL. See Azure AI Foundry Setup for detailed instructions |
| Ollama | Connects to a locally-running Ollama instance. Set the Base URL to the Ollama host (defaults to http://localhost:11434) |
| OpenAI-Compatible | Connects to any API that follows the OpenAI API format. Requires a Base URL |
| Amazon Bedrock | Connects to AWS Bedrock foundation models |
| Groq | High-speed inference service for open-source models |
| Mistral | Connects to Mistral AI models |
| Cohere | Connects to Cohere's language and embedding models |
| Perplexity | Connects to Perplexity's search-augmented models |
| DeepSeek | Connects to DeepSeek's reasoning and coding models |
| OpenRouter | Multi-provider routing service with access to many model providers |
Managing Models
After creating a provider, open the Models tab to control which models are available to your organization.
When the Models tab loads, Raikoo automatically queries the provider's API to discover available models and matches them against its built-in library of known model definitions.
Model Management Modes
Use the toggle at the top of the Models tab to choose between two modes:
Allow All
All discovered models from the provider are available by default. You can individually exclude specific models you do not want users to access. When new models appear from the provider's API, they become available automatically.
Select Models
No models are available by default. You must explicitly enable each model you want to expose. New models from the provider API do not appear automatically -- use Refresh to check for additions and enable them as needed.
Which mode should I use?
Use Allow All when you trust the provider's full model list and want new models to appear automatically. Use Select Models when you need precise control over which models users can access.
Model Discovery
Each discovered model is shown in a list with the following information:
- Model ID -- The identifier returned by the provider's API
- Name -- A human-readable name, if resolved from a known definition
- Creator -- The model's creator (e.g., OpenAI, Google) or "Custom" for manually added models
- Resolution confidence -- How the model was matched to a known definition:
| Badge | Meaning |
|---|---|
| Matched (green) | Model ID exactly matched a known Raikoo definition |
| Inferred (yellow) | Model ID was matched by pattern inference (e.g., a deployment name containing a known model stem) |
| Unknown (grey) | No match found; no pre-loaded capability data |
- Capability confidence -- How complete the capability data is:
| Badge | Meaning |
|---|---|
| Full info (green) | Context window, capabilities, and modalities are all defined |
| Partial info (yellow) | Some capability fields are defined |
| No info (grey) | Only the model ID is known |
- Context window and max output -- Token limits, if known
- Modality icons -- Input and output types (text, image, audio, video)
- Capability chips -- Quick indicators for tools, system prompt, and reasoning support
Click Refresh at any time to re-query the provider's API for updated model information.
Editing Model Capabilities
Click any model row to open the model editor. Here you can view and customize the model's capability data.
Model Preset
The preset dropdown lets you select a known model from Raikoo's built-in library to apply its full capability profile. This is useful when a provider returns a custom deployment name that Raikoo did not automatically match -- for example, selecting the gpt-4o preset for a deployment named my-company-gpt4o-prod.
Capability Fields
| Field | Description |
|---|---|
| Name | Display name shown in model selectors throughout the UI |
| Description | Descriptive text about this model's purpose and characteristics |
| Context window | Maximum tokens this model can process per request (input and output combined) |
| Max output tokens | Maximum tokens this model can generate per response |
| Input modalities | Content types the model accepts: text, image, audio, video |
| Output modalities | Content types the model produces: text, image, audio, video, embeddings |
| System prompt | Whether this model supports a system or developer prompt |
| Tools | Whether this model supports function calling and tool use |
| Reasoning effort levels | Comma-separated levels for extended thinking (e.g., low,medium,high) |
| Max budget tokens | Maximum token budget for reasoning |
Why accuracy matters
Accurate capability data lets Raikoo make better decisions. For example, image attachment controls only appear for models that support vision input, and reasoning effort controls are only shown for models that support extended thinking. Incorrect data may cause features to appear or be hidden unexpectedly.
Adding a Custom Model ID
In Select Models mode, you can add model IDs that are not in the discovered list. This is useful for:
- Provider-specific deployment names that do not appear in the discovery API
- Models in preview or early access that are not yet listed
- Custom fine-tuned models
To add a custom model:
- Switch to Select Models mode
- Click Add Model in the toolbar
- Enter the model ID exactly as the provider expects it
- Click Add -- the model editor opens so you can configure its capabilities
Custom model IDs
Custom model IDs are not validated against the provider's API. Ensure the ID matches exactly what the provider expects, or API calls using this model will fail.
Model Families
Model families are logical groupings of models with an ordered fallback sequence. When a workflow or agent targets a family, Raikoo automatically selects the best available model from the group.
Model families are managed from Model Families in the left sidebar, under the same Configuration section as AI Providers.
To create a model family:
- Navigate to Model Families and click Create
- Enter a name for the family (e.g., "Fast Models" or "High Capability")
- Add models in priority order -- the first available model in the list will be used
- Save the family
For a conceptual explanation of model families and when to use them, see Models.
Permissions Required
You need the model-families organization permission to manage model families. See Roles & Permissions for details.
Azure AI Foundry
For detailed instructions on connecting to Azure AI Foundry and Azure OpenAI deployments, see the dedicated Azure AI Foundry Setup guide.
Troubleshooting
No Models Discovered
Possible causes:
- The API key is incorrect or has expired
- The Base URL is wrong or missing (required for Azure OpenAI, Ollama, and OpenAI-Compatible providers)
- The provider type does not support model discovery
Solutions:
- Verify the API key is correct and has not been revoked
- Check the Base URL matches exactly what the provider expects
- For Azure OpenAI, see Azure AI Foundry Setup for the correct URL format
- If discovery is not supported by your provider, add models manually using Add Model in Select Models mode
API Key Errors
Symptoms: Discovery fails or API calls return authentication errors.
Solutions:
- Re-enter the API key on the Configuration tab. For security, existing keys are not displayed after saving
- Generate a new API key from the provider's dashboard
- Verify the key has the required permissions for model listing and chat completions
Models Missing After Refresh
Symptoms: A model you expect to see does not appear after clicking Refresh.
Solutions:
- Allow a few minutes for new models to propagate through the provider's API
- Confirm the model is available in your provider account or subscription tier
- Check if the model was excluded (in Allow All mode) or not selected (in Select Models mode)
Next Steps
- Understand the model hierarchy: Models
- Configure agents to use specific models: Agents
- Use models in chat sessions: Chat
- Set up Azure AI Foundry: Azure AI Foundry Setup