# Run API Streaming Example
This example shows how to stream tokens directly from `/<pipeline>/run` endpoint
using `streaming_generator()`. Instead of waiting for the pipeline to finish and returning
the final string, the wrapper yields streaming chunks as soon as the underlying LLM produces them.
## Highlights
- Implements `run_api()` to return a generator of `StreamingChunk` objects.
- Uses the helper `streaming_generator()` from Hayhooks (no manual queue management required).
- Demonstrates how `/run` automatically becomes a `StreamingResponse` with `text/plain` media type.
- Works seamlessly with curl, CLI tools, and any HTTP client.
## Deploy & Try It
```bash
# 1. Set your OpenAI API key (the pipeline uses OpenAIChatGenerator)
export OPENAI_API_KEY=your_api_key_here
# 2. Deploy the example
hayhooks deploy examples/pipeline_wrappers/run_api_streaming
# 3. Call the /run endpoint with streaming enabled (-N keeps curl open)
curl -N http://localhost:1416/run_api_streaming/run \
-H "Content-Type: application/json" \
-d '{"question": "What is Redis?", "urls": ["https://www.redis.io"]}'
```
You should see words streaming in the terminal instead of receiving the entire answer at once.
## How It Works
```python
def run_api(self, question: str, urls: list[str] | None = None) -> Generator[StreamingChunk, None, None]:
return streaming_generator(
pipeline=self.pipeline,
pipeline_run_args={
"fetcher": {"urls": urls or DEFAULT_URLS},
"prompt": {"query": question},
},
)
```
By returning the generator, Hayhooks automatically wraps the response in a `StreamingResponse`
and takes care of cleaning up the generator once streaming is complete.
> **Note:** For async pipelines, use `async_streaming_generator()` inside `run_api_async()` instead.
> The async version works identically but returns an async generator that Hayhooks will handle automatically.
## Alternative: Server-Sent Events (SSE)
If you need SSE format (e.g., for browser `EventSource` or clients expecting `text/event-stream`),
wrap the generator with `SSEStream`:
```python
from hayhooks import SSEStream, streaming_generator
def run_api(self, question: str, urls: list[str] | None = None):
return SSEStream(
streaming_generator(
pipeline=self.pipeline,
pipeline_run_args={
"fetcher": {"urls": urls or DEFAULT_URLS},
"prompt": {"query": question},
},
)
)
```
This changes the response media type to `text/event-stream` and wraps each chunk in SSE format.
