> ## Documentation Index
> Fetch the complete documentation index at: https://docs.welesai.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Configure AI providers

> Connect AI model providers to Weles (Ollama, OpenAI, GroqCloud, DeepSeek, Gemini, NVIDIA NIM) and select default LLM, embedding, and OCR models.

Weles separates provider connections from the model selected for a document type or agent stage. Configure providers as an administrator in **LLM Config** before creating workflows.

## Add a provider

1. Open **LLM Config** and select **Add Provider** under **Configured Providers**.
2. Select **Provider Type**: **Ollama (Local)**, **OpenAI**, **GroqCloud**, **DeepSeek**, **Google (Gemini)**, or **NVIDIA NIM**. This chooses the API format. An OpenAI-compatible endpoint can use the OpenAI option when its API is compatible; test the actual service and model.
3. Optionally enter a **Display Name** to distinguish connections. Choose **Type**: **LLM Only**, **Embedding Only**, or **LLM + Embedding**. This controls which active-provider selectors show the connection; it does not add capabilities the service lacks.
4. Enter the **API Key** if required, the **Base URL** reachable from the Weles containers, and an available **Default Model** identifier. Selecting a provider type can prefill defaults; replace them with the values approved for your deployment. Model names and service capabilities can change.
5. Select **Save**, then use the provider card's connection test. Use the model list/refresh control to inspect available models. Confirm the desired model is available for your account and supports the intended task, including tool calls or image input when needed.
6. In **Active Providers**, select the connection for **LLM Provider**, **Embedding Provider**, and **OCR Provider** as needed. The selection is saved when changed. Only providers marked LLM or both appear for LLM and OCR; providers marked embedding or both appear for embeddings.

When no active LLM or embedding provider is selected, Weles falls back to Ollama. With no OCR provider selected, it uses the active LLM provider. A reachable provider and compatible model are still required.

## Example: local Ollama

Choose **Ollama (Local)**, provide a recognizable display name and the Ollama base URL visible to the backend and worker containers, and enter the name of a model already available on that Ollama server. Save, test the connection, and choose it as the active LLM provider. The deployment's `OLLAMA_HOST` is used as the default Ollama URL; verify host networking on your platform. `localhost` inside a container refers to that container, not the host. For embeddings, configure a suitable embedding model and verify its compatibility with the collection.

## Example: OpenAI

Choose **OpenAI**, enter the service API key, review the prefilled base URL, and enter an available model ID for your account. Choose **LLM Only** unless you have verified that the selected model and provider configuration are appropriate for embeddings too. Save, test, and select the provider as the active LLM. For embeddings, consider a separate provider entry with an embedding model as its default.

## Choosing models for workflows

A document type or agent stage can specify a model name. The agent stage uses the active LLM provider but passes its configured model name to that provider. In the current builder the **Model Ollama** list is populated by the active LLM provider's model-list endpoint; despite the label it is not a separate provider selector. Ensure the chosen model ID works with the active provider. Test agent tool calling and any vision requirements with the actual model.

> [!WARNING]
>
> External providers can receive prompts, document content, and tool results. Review data-handling terms and your organization's policies. Provider API keys are stored in the application database; do not assume the Secrets Vault encrypts provider keys. Protect database access and backups.

If a test or model listing fails, check the endpoint from the container network, API key, TLS, firewall rules, provider account permissions, and model name. A connection test alone does not prove that the configured model supports tool calls, OCR, or embeddings. See [agent workflows](https://docs.welesai.com/workflows/agent-workflow.md) and [troubleshooting](https://docs.welesai.com/administration/troubleshooting.md).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.