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

# Azure AI Foundry

> Use open-source models (Phi, Llama, Mistral, Cohere) hosted on Azure AI Foundry with Agentium.

# Azure AI Foundry

Use open-source and partner models hosted on **Azure AI Foundry** (formerly Azure AI Model Catalog) — access Phi, Llama, Mistral, Cohere, and more through Azure's infrastructure with enterprise compliance.

Azure AI Foundry exposes an OpenAI-compatible API, so Agentium connects to it using the standard `openai` SDK pointed at your Azure endpoint.

***

## Setup

<Tabs>
  <Tab title="Install">
    Uses the standard OpenAI SDK:

    ```bash theme={null}
    npm install openai
    ```
  </Tab>

  <Tab title="Environment">
    Set your Azure AI Foundry credentials:

    ```bash theme={null}
    export AZURE_API_KEY="..."
    export AZURE_ENDPOINT="https://<your-host>.<region>.models.ai.azure.com"
    ```

    Navigate to [Azure AI Foundry](https://ai.azure.com) on the Azure Portal to create a service and get your endpoint URL.
  </Tab>
</Tabs>

***

## Factory

```typescript theme={null}
import { azureFoundry } from "@agentium/core";

const model = azureFoundry("Phi-4");
```

<ParamField path="modelId" type="string" required>
  The model name as deployed on Azure AI Foundry (e.g., `"Phi-4"`, `"Meta-Llama-3.1-70B-Instruct"`).
</ParamField>

<ParamField path="config" type="AzureFoundryConfig" required={false}>
  Optional configuration. See Config below.
</ParamField>

***

## Supported Models

### Microsoft Models

| Model                     | Description                                      |
| ------------------------- | ------------------------------------------------ |
| `Phi-4`                   | Latest Phi model. Strong reasoning for its size. |
| `Phi-3.5-mini-instruct`   | Fast, compact, good for edge scenarios.          |
| `Phi-3.5-vision-instruct` | Vision-capable Phi model.                        |

### Meta Models

| Model                          | Description                                 |
| ------------------------------ | ------------------------------------------- |
| `Meta-Llama-3.1-405B-Instruct` | Largest open model. Exceptional capability. |
| `Meta-Llama-3.1-70B-Instruct`  | Strong balance of size and performance.     |
| `Meta-Llama-3.1-8B-Instruct`   | Fast, efficient, good for most tasks.       |

### Mistral Models

| Model           | Description                        |
| --------------- | ---------------------------------- |
| `Mistral-large` | High capability, strong reasoning. |
| `Mistral-small` | Fast, cost-effective.              |
| `Mistral-Nemo`  | Latest Nemo model.                 |

### Cohere Models

| Model                   | Description              |
| ----------------------- | ------------------------ |
| `Cohere-command-r-plus` | Strong RAG and tool use. |
| `Cohere-command-r`      | Fast, efficient.         |

<Accordion title="Discovering more models">
  Browse the full model catalog at [Azure AI Foundry](https://ai.azure.com/explore/models). Available models depend on your Azure region and subscription.
</Accordion>

***

## Config

<ParamField path="apiKey" type="string" required={false}>
  Azure API key. Falls back to `AZURE_API_KEY` env var.
</ParamField>

<ParamField path="endpoint" type="string" required>
  Azure AI Foundry endpoint URL. Falls back to `AZURE_ENDPOINT` env var. Format: `https://<host>.<region>.models.ai.azure.com`
</ParamField>

<ParamField path="apiVersion" type="string" required={false}>
  API version. Falls back to `AZURE_API_VERSION` env var.
</ParamField>

***

## Authentication

### Environment Variables (Recommended)

```bash theme={null}
export AZURE_API_KEY="..."
export AZURE_ENDPOINT="https://my-host.eastus.models.ai.azure.com"
```

```typescript theme={null}
const model = azureFoundry("Phi-4");
```

### Explicit Config

```typescript theme={null}
const model = azureFoundry("Phi-4", {
  apiKey: "...",
  endpoint: "https://my-host.eastus.models.ai.azure.com",
});
```

***

## Tool Calling

Tool calling is supported by models that implement function calling (Mistral Large, Llama 3.1, Cohere Command R/R+):

```typescript theme={null}
import { Agent, azureFoundry, defineTool } from "@agentium/core";
import { z } from "zod";

const agent = new Agent({
  name: "foundry-assistant",
  model: azureFoundry("Mistral-large"),
  instructions: "You are a helpful assistant with tool access.",
  tools: [
    defineTool({
      name: "getStockPrice",
      description: "Get current stock price",
      parameters: z.object({ symbol: z.string() }),
      execute: async ({ symbol }) => `${symbol}: $185.42 (+2.3%)`,
    }),
  ],
});

const result = await agent.run("What's the current price of AAPL?");
console.log(result.text);
```

<Warning>
  Not all Azure AI Foundry models support tool calling. Phi models and some smaller models may not support function calling. Test with your specific model.
</Warning>

***

## Vision Models

Some Azure AI Foundry models support image input (e.g., `Phi-3.5-vision-instruct`):

```typescript theme={null}
import { Agent, azureFoundry, type ContentPart } from "@agentium/core";
import { readFileSync } from "node:fs";

const agent = new Agent({
  name: "vision-agent",
  model: azureFoundry("Phi-3.5-vision-instruct"),
  instructions: "Describe images in detail.",
});

const imageData = readFileSync("photo.jpg").toString("base64");
const result = await agent.run([
  { type: "text", text: "What's in this image?" },
  { type: "image", data: imageData, mimeType: "image/jpeg" },
] as ContentPart[]);
```

***

## Full Example

```typescript theme={null}
import { Agent, azureFoundry, CostTracker, defineTool } from "@agentium/core";
import { z } from "zod";

const costTracker = new CostTracker({
  pricing: {
    "Meta-Llama-3.1-70B-Instruct": { promptPer1k: 0.00268, completionPer1k: 0.00354 },
    "Mistral-large": { promptPer1k: 0.004, completionPer1k: 0.012 },
  },
});

const agent = new Agent({
  name: "research-agent",
  model: azureFoundry("Meta-Llama-3.1-70B-Instruct", {
    endpoint: "https://my-host.eastus.models.ai.azure.com",
  }),
  instructions: "You are a research assistant. Use tools to find information.",
  tools: [
    defineTool({
      name: "search",
      description: "Search for information",
      parameters: z.object({ query: z.string() }),
      execute: async ({ query }) => `Results for "${query}": ...`,
    }),
  ],
  costTracker,
  maxToolRoundtrips: 3,
});

const result = await agent.run("Research the latest trends in AI agent frameworks");
console.log(result.text);
console.log(`Cost: $${costTracker.getSummary().totalCost.toFixed(4)}`);
```

***

## Azure AI Foundry vs Other Providers

| Feature      | `openai()`   | `azureFoundry()`            | `ollama()`           |
| ------------ | ------------ | --------------------------- | -------------------- |
| Models       | OpenAI only  | Phi, Llama, Mistral, Cohere | Any local model      |
| Auth         | API key      | Azure API key               | None                 |
| Hosting      | OpenAI cloud | Azure cloud                 | Local machine        |
| Cost         | Per-token    | Per-token (Azure pricing)   | Free (your hardware) |
| Privacy      | OpenAI terms | Azure data residency        | Fully local          |
| Compliance   | SOC2         | SOC2, HIPAA, GDPR           | N/A                  |
| GPU required | No           | No                          | Yes (recommended)    |

Use `azureFoundry()` when you want open-source model quality with Azure enterprise compliance — without managing GPU infrastructure.

***

## Environment Variables

| Variable            | Description                   |
| ------------------- | ----------------------------- |
| `AZURE_API_KEY`     | Azure AI Foundry API key      |
| `AZURE_ENDPOINT`    | Azure AI Foundry endpoint URL |
| `AZURE_API_VERSION` | API version (optional)        |

***

## Cross-References

* [Azure OpenAI](/models/azure-openai) — OpenAI GPT/o-series models on Azure
* [Ollama](/models/ollama) — Run the same open-source models locally
* [AWS Bedrock](/models/aws-bedrock) — Similar model catalog on AWS
* [Custom Provider](/models/custom-provider) — Build your own provider