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

# LLM Providers

> Connect your AI endpoint. Supported providers, configuration, and security notes.

# LLM Providers

TestForge uses a **Bring Your Own Key (BYOK)** model. You connect your own LLM endpoint — TestForge never processes your data on its servers.

## Supported providers

| Provider                 | Type             | Data residency               |
| ------------------------ | ---------------- | ---------------------------- |
| **Azure AI Foundry**     | Browser-direct   | ✅ Stays in your Azure tenant |
| **Azure OpenAI Service** | Browser-direct   | ✅ Stays in your Azure tenant |
| **OpenAI**               | Browser-direct   | ⚠️ Leaves your tenant        |
| **Mistral**              | Browser-direct   | ⚠️ Leaves your tenant        |
| **Gemini**               | Browser-direct   | ⚠️ Leaves your tenant        |
| **Anthropic**            | Via client proxy | ⚠️ Leaves your tenant        |

<Warning>
  For OpenAI, Mistral, Gemini, and Anthropic, your user story content is sent to an external API outside your Azure tenant. TestForge displays a non-dismissible warning before saving.
</Warning>

## Azure AI Foundry (recommended)

Endpoint format: `https://<your-resource>.cognitiveservices.azure.com`

Compatible models: `gpt-4.1-mini`, `gpt-4.1`, `gpt-4o`, `gpt-4o-mini`.

<Note>
  Models using the Responses API (`gpt-5.1-codex-*`, `o3`, `o4-mini`) are not yet supported. Use `gpt-4.1-mini` for best results.
</Note>

## Azure OpenAI Service

Endpoint format: `https://<your-resource>.openai.azure.com`

Configuration follows the same steps as Azure AI Foundry.

## Anthropic (via client proxy)

Anthropic blocks browser-direct CORS calls. Deploy the open-source proxy template in your tenant (Azure Function or Cloudflare Worker) and enter your proxy URL in the Configuration tab.

## Token cost estimate

A typical session (score + analysis + 9 manual tests + Playwright code) consumes approximately **15,000–25,000 tokens**. At `gpt-4.1-mini` pricing on Azure, this is roughly **€0.05–0.10 per session**.