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

# Model providers

> Configure supported LLM and embedding providers, understand capability routing, defaults, dimensions, and fallback behavior.

InteraOne selects LLM and embedding providers independently. This allows, for example, a local Ollama chat model with hosted embeddings or Bedrock generation with OpenAI embeddings.

## Supported implementations

| Provider     | LLM setting            | Embedding setting            | Default model in source               |
| ------------ | ---------------------- | ---------------------------- | ------------------------------------- |
| AWS Bedrock  | `LLM_PROVIDER=bedrock` | `EMBEDDING_PROVIDER=bedrock` | Nova Pro / Titan Embed Text v2        |
| Hugging Face | `huggingface`          | `huggingface`                | Llama 3.3 70B Instruct / MiniLM L6 v2 |
| Ollama       | `ollama`               | `ollama`                     | Llama 3.2 / nomic-embed-text          |
| OpenAI       | `openai`               | `openai`                     | GPT-4o mini / text-embedding-3-small  |

<Note>
  The agent config also contains Gemini fields, but the provider factory source currently ships concrete Bedrock, Hugging Face, Ollama, and OpenAI implementations. Do not advertise or select a provider until both factory registration and implementation exist.
</Note>

## Configure a provider

<Tabs>
  <Tab title="OpenAI">
    ```bash theme={null}
    LLM_PROVIDER=openai
    OPENAI_API_KEY=...
    OPENAI_MODEL=gpt-4o-mini
    EMBEDDING_PROVIDER=openai
    OPENAI_EMBEDDING_MODEL=text-embedding-3-small
    OPENAI_EMBEDDING_DIMENSIONS=1536
    ```
  </Tab>

  <Tab title="Bedrock">
    ```bash theme={null}
    LLM_PROVIDER=bedrock
    AWS_REGION=us-east-1
    AWS_ACCESS_KEY_ID=...
    AWS_SECRET_ACCESS_KEY=...
    BEDROCK_MODEL=us.amazon.nova-pro-v1:0
    EMBEDDING_PROVIDER=bedrock
    BEDROCK_EMBEDDING_MODEL=amazon.titan-embed-text-v2:0
    BEDROCK_EMBEDDING_DIMENSIONS=1024
    ```
  </Tab>

  <Tab title="Ollama">
    ```bash theme={null}
    LLM_PROVIDER=ollama
    OLLAMA_HOST=ollama
    OLLAMA_PORT=11434
    OLLAMA_MODEL=llama3.2
    EMBEDDING_PROVIDER=ollama
    OLLAMA_EMBEDDING_MODEL=nomic-embed-text
    OLLAMA_EMBEDDING_DIMENSIONS=768
    ```
  </Tab>

  <Tab title="Hugging Face">
    ```bash theme={null}
    LLM_PROVIDER=huggingface
    HF_TOKEN=...
    HF_MODEL=meta-llama/Llama-3.3-70B-Instruct
    EMBEDDING_PROVIDER=huggingface
    HF_EMBEDDING_MODEL=sentence-transformers/all-MiniLM-L6-v2
    HF_EMBEDDING_DIMENSIONS=384
    ```
  </Tab>
</Tabs>

## Capability and health routing

The model registry describes streaming and tool support instead of inferring behavior from model-name strings. Routing code records model health, uses provider-specific implementations behind common interfaces, and can apply fallbacks.

<Warning>
  Test tool use, streaming, token accounting, and context limits for the exact model identifier you deploy. Provider support does not imply every model from that provider has identical capabilities.
</Warning>