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

# Providers

> OpenAI, Anthropic, Google, and Amazon Bedrock provider setup and configuration

The Python SDK uses [LiteLLM](https://docs.litellm.ai/) under the hood for unified provider access. LiteLLM is included as a dependency, so no additional provider packages are needed.

## Installation

```bash theme={null}
pip install tracia
```

LiteLLM is installed automatically with Tracia and handles all provider communication.

## OpenAI

### Environment Variable

```bash theme={null}
OPENAI_API_KEY=sk-your-openai-key
```

### Usage

```python theme={null}
result = client.run_local(
    model="gpt-4o",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Hello!"},
    ],
    temperature=0.7,
    max_output_tokens=500,
)
```

### Streaming

```python theme={null}
stream = client.run_local(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Write a poem."}],
    stream=True,
)

for chunk in stream:
    print(chunk, end="")
```

## Anthropic

### Environment Variable

```bash theme={null}
ANTHROPIC_API_KEY=sk-ant-your-anthropic-key
```

### Usage

```python theme={null}
result = client.run_local(
    model="claude-sonnet-4-20250514",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Explain quantum computing."},
    ],
    temperature=0.7,
    max_output_tokens=1000,
)
```

### Streaming

```python theme={null}
stream = client.run_local(
    model="claude-sonnet-4-20250514",
    messages=[{"role": "user", "content": "Write a story."}],
    stream=True,
)

for chunk in stream:
    print(chunk, end="")
```

## Google

### Environment Variable

```bash theme={null}
GOOGLE_API_KEY=your-google-api-key
```

### Usage

```python theme={null}
result = client.run_local(
    model="gemini-2.0-flash",
    messages=[
        {"role": "user", "content": "What are the benefits of Python?"},
    ],
    temperature=0.5,
    max_output_tokens=800,
)
```

### Streaming

```python theme={null}
stream = client.run_local(
    model="gemini-2.0-flash",
    messages=[{"role": "user", "content": "Explain machine learning."}],
    stream=True,
)

for chunk in stream:
    print(chunk, end="")
```

## Provider Override

The SDK automatically detects the provider based on the model name. For custom or fine-tuned models not in the built-in list, specify the provider explicitly:

```python theme={null}
# Fine-tuned OpenAI model
result = client.run_local(
    model="ft:gpt-4o:my-org:custom-model:abc123",
    provider="openai",
    messages=[{"role": "user", "content": "Hello!"}],
)
```

## API Key Override

Override the environment variable for a specific request:

```python theme={null}
import os

result = client.run_local(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Hello!"}],
    provider_api_key=os.environ["OPENAI_API_KEY_SECONDARY"],
)
```

This is useful for:

* Multi-tenant applications with per-customer API keys
* Testing with different API keys
* Using organization-specific keys

## Error Handling by Provider

```python theme={null}
from tracia import Tracia, TraciaError, TraciaErrorCode

try:
    result = client.run_local(
        model="gpt-4o",
        messages=[{"role": "user", "content": "Hello!"}],
    )
except TraciaError as error:
    if error.code == TraciaErrorCode.MISSING_PROVIDER_API_KEY:
        print("Set OPENAI_API_KEY environment variable")
    elif error.code == TraciaErrorCode.PROVIDER_ERROR:
        print(f"Provider API error: {error.message}")
    elif error.code == TraciaErrorCode.UNSUPPORTED_MODEL:
        print("Specify provider explicitly for custom models")
```

## Provider Comparison

| Feature           | OpenAI          | Anthropic       | Google               | Amazon Bedrock   |
| ----------------- | --------------- | --------------- | -------------------- | ---------------- |
| System messages   | Native support  | Native support  | Converted internally | Vendor-dependent |
| Temperature range | 0-2             | 0-1             | 0-2                  | Vendor-dependent |
| Max output tokens | Model-dependent | Model-dependent | Model-dependent      | Model-dependent  |
| Stop sequences    | Supported       | Supported       | Supported            | Supported        |
| Streaming         | Supported       | Supported       | Supported            | Supported        |
| Tool calling      | Supported       | Supported       | Supported            | Model-dependent  |

<Note>
  Amazon Bedrock hosts models from multiple vendors (Amazon, Anthropic, Meta, Mistral, etc.). Feature support depends on the specific model. See the [Bedrock page](/sdk-python/run-local/bedrock) for details.
</Note>