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

# Configure the WeCareRemote AI assistant

> Customize your AI assistant experience by choosing a model, setting up API authentication, and connecting your preferred LLM provider on WeCareRemote.

The WeCareRemote AI assistant gives you control over which AI model powers your conversations and how you authenticate API requests. This page explains the configuration options available to you as a user or administrator.

Most users will not need to change anything — your NGO administrator sets sensible defaults when the platform is provisioned. This page is most useful for administrators configuring a new deployment and for developers calling the API directly.

## Choosing your AI model

WeCareRemote supports a wide range of AI models from multiple providers. Your administrator sets the platform default, but you can override the model on individual API requests by passing the `model` field in your request body.

```json theme={null}
{
  "message": "What housing support is available?",
  "model": "claude-3-5-sonnet-20241022"
}
```

If no model is specified, the assistant uses the platform default configured for your organization. To see all available models, visit [AI models supported by WeCareRemote](/ai-assistant/supported-models).

<Tip>
  For most use cases, GPT-4o and Claude 3.5 Sonnet deliver the best balance of quality and speed. If your organization requires data to remain on-premises, ask your administrator about enabling a locally hosted model via Ollama.
</Tip>

## API authentication

When accessing the AI assistant through the REST API, every request must include a Bearer token in the `Authorization` header.

```http theme={null}
Authorization: Bearer <your-token>
```

You obtain this token by logging in to WeCareRemote. See [Authenticate API requests](/api/authentication) for a step-by-step guide to getting and using your token.

<Warning>
  Keep your Bearer token secure. Never share it publicly, commit it to version control, or expose it in client-side code. Contact your WeCareRemote administrator if you believe your token has been compromised.
</Warning>

## Supported LLM providers

WeCareRemote connects to the following AI providers. Your organization's administrator controls which providers are active on your instance.

<AccordionGroup>
  <Accordion title="OpenAI">
    Provides GPT-4o, GPT-4o-mini, GPT-4 Turbo, and GPT-3.5 Turbo. Well-suited for general conversation, summarization, and document Q\&A.
  </Accordion>

  <Accordion title="Anthropic">
    Provides Claude 3.5 Sonnet, Claude 3 Haiku, and Claude 3 Opus. Known for nuanced, instruction-following responses.
  </Accordion>

  <Accordion title="Google">
    Provides Gemini 1.5 Pro and Gemini 1.5 Flash via Google AI Studio or Vertex AI. Strong multimodal and long-context performance.
  </Accordion>

  <Accordion title="Groq">
    Provides fast inference for open-weight models including Llama 3.1 and Mixtral. Ideal for low-latency interactions.
  </Accordion>

  <Accordion title="Ollama (local)">
    Runs open-weight models locally on your organization's own hardware. No data leaves your environment — the recommended option for privacy-sensitive use cases.
  </Accordion>

  <Accordion title="AWS Bedrock">
    Provides managed access to Anthropic Claude, Meta Llama, Amazon Titan, and other models through Amazon's infrastructure.
  </Accordion>

  <Accordion title="Azure OpenAI">
    Provides GPT models through your organization's Azure OpenAI resource, with enterprise-grade compliance and data residency controls.
  </Accordion>

  <Accordion title="OpenRouter">
    Aggregates models from many providers — including OpenAI, Anthropic, Google, Meta, and Mistral — through a single API.
  </Accordion>

  <Accordion title="DeepSeek">
    Provides capable reasoning and coding-focused models via the DeepSeek API.
  </Accordion>

  <Accordion title="Custom OpenAI-compatible endpoints">
    Connect any API that follows the OpenAI API format — including self-hosted models, fine-tuned endpoints, or third-party providers not listed above.
  </Accordion>
</AccordionGroup>

## Conversation history

The assistant stores your conversation history so you can pick up where you left off across sessions. Your organization's administrator determines how long history is retained and which storage backend is used. If you have questions about data retention, contact your WeCareRemote administrator.

For more details on how memory and history work, see [Conversation memory and chat history](/ai-assistant/memory-and-history).

## Choosing a provider

If you are an administrator setting up the platform for the first time, consider these trade-offs:

| Priority                    | Recommended providers                          |
| --------------------------- | ---------------------------------------------- |
| Best general quality        | OpenAI (GPT-4o), Anthropic (Claude 3.5 Sonnet) |
| Lowest latency              | Groq, GPT-4o-mini                              |
| Lowest cost                 | Groq, GPT-4o-mini, OpenRouter                  |
| Data residency / compliance | Azure OpenAI, AWS Bedrock                      |
| Full on-premises / private  | Ollama, custom OpenAI-compatible endpoint      |
| Strong multilingual         | Claude 3.5 Sonnet, Gemini 1.5 Pro, GPT-4o      |

You can mix and match: set a high-quality model as the default and let advanced users override the model per request when they need something faster or cheaper.

## Frequently asked questions

<AccordionGroup>
  <Accordion title="Can refugee users switch the model themselves?">
    Through the web UI, no — the model is set by your administrator. The per-request override is only available when calling the API directly with a `model` parameter.
  </Accordion>

  <Accordion title="Does switching models lose the conversation?">
    No. Conversation history is stored independently of the model. You can switch models mid-conversation and the assistant will still have full context.
  </Accordion>

  <Accordion title="Can I use my organization's own API key?">
    Yes. Administrators configure each provider's API key when enabling that provider on the platform. Contact your administrator if you want to bring your own key.
  </Accordion>

  <Accordion title="What happens if the configured model is unavailable?">
    If the provider returns an error, the API surfaces a `500` response. Administrators can configure a fallback provider in the platform settings.
  </Accordion>
</AccordionGroup>