> For the complete documentation index, see [llms.txt](https://opengpu-network.gitbook.io/opengpu-network/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://opengpu-network.gitbook.io/opengpu-network/for-clients/using-the-sdk.md).

# Using the SDK

The OpenGPU Python SDK provides programmatic access to the network for automation and integration. Currently Python-only, covering core functionality.

**Full documentation**: [opengpu-network.github.io/sdk-ogpu-py](https://opengpu-network.github.io/sdk-ogpu-py/)

## Installation

```bash
pip install ogpu
```

## Configuration

Create a `.env` file with your credentials:

```
CLIENT_PRIVATE_KEY=your_private_key_here
```

## Verify Installation

```python
import ogpu
print(f"OpenGPU SDK version: {ogpu.__version__}")

from ogpu.service import task
from ogpu.client import OGPUClient
print("OpenGPU SDK installed successfully!")
```

## Client API

### Publish a Source

```python
from ogpu.client import publish_source, SourceInfo, ImageEnvironments, DeliveryMethod

source_info = SourceInfo(
    name="My-Model",
    description="My AI model",
    image_environments=ImageEnvironments(
        nvidia="docker-compose-nvidia.yml"
    ),
    delivery_method=DeliveryMethod.FIRST_RESPONSE
)

source_address = publish_source(source_info)
```

### Publish a Task

```python
from ogpu.client import publish_task, TaskInfo, TaskInput

task_info = TaskInfo(
    source_address="0x...",
    task_input=TaskInput(
        function_name="process",
        input_data={"text": "Hello!"}
    ),
    expiry_time=3600,
    payment=0.01
)

task_address = publish_task(task_info)
```

### Get Responses

```python
from ogpu.client import get_task_responses, confirm_response

responses = get_task_responses(task_address)

# For Manual Confirmation delivery method
confirm_response(response_address)
```

## Service API

For building task handlers that run on provider nodes:

```python
import ogpu.service
from pydantic import BaseModel

class Request(BaseModel):
    text: str

class Response(BaseModel):
    result: str

@ogpu.service.init()
def load_model():
    ogpu.service.logger.info("Loading model...")

@ogpu.service.expose(timeout=60)
def process(data: Request) -> Response:
    return Response(result="processed")

if __name__ == "__main__":
    ogpu.service.start()
```

## Network Selection

```python
from ogpu.client import set_chain, ChainId

set_chain(ChainId.OGPU_MAINNET)  # Chain ID: 1071
# or
set_chain(ChainId.OGPU_TESTNET)  # Chain ID: 200820172034
```

## Best Practices

* **Store private keys securely**: Use environment variables
* **Set appropriate timeouts**: Network operations may take time
* **Use Pydantic models**: Required for service input/output

> See [SDK Documentation](https://opengpu-network.github.io/sdk-ogpu-py/) for API reference.