Streaming

Get real-time feedback during extraction for better UX.

---

Why Streaming?

For large documents, extraction can take 10-30 seconds. Streaming provides:

• Progress feedback — Users see something is happening
• Early results — Partial data before completion
• Cancellation — Stop extraction mid-way

---

Basic Streaming

Use the stream parameter:

from strutex import DocumentProcessor, GeminiProvider

processor = DocumentProcessor(provider=GeminiProvider())

# Stream extraction
for chunk in processor.stream(
    file_path="large_document.pdf",
    prompt="Extract all sections",
    schema=MySchema
):
    print(chunk, end="", flush=True)

---

Streaming with Callbacks

Get structured updates during extraction:

from strutex import DocumentProcessor, GeminiProvider

def on_chunk(chunk: str, context: dict):
    """Called for each streamed chunk."""
    print(f"Received: {chunk[:50]}...")

def on_complete(result: dict, context: dict):
    """Called when extraction completes."""
    print(f"Complete! Got {len(result)} fields")

processor = DocumentProcessor(provider=GeminiProvider())

result = processor.stream(
    file_path="document.pdf",
    prompt="Extract data",
    schema=MySchema,
    on_chunk=on_chunk,
    on_complete=on_complete
)

---

Streaming in Web Apps

FastAPI with SSE

from fastapi import FastAPI
from fastapi.responses import StreamingResponse
from strutex import DocumentProcessor, GeminiProvider

app = FastAPI()
processor = DocumentProcessor(provider=GeminiProvider())

@app.get("/extract/stream")
async def stream_extraction(file_path: str):
    async def generate():
        for chunk in processor.stream(file_path, "Extract", schema=MySchema):
            yield f"data: {chunk}\n\n"
        yield "data: [DONE]\n\n"

    return StreamingResponse(
        generate(),
        media_type="text/event-stream"
    )

JavaScript Client

const eventSource = new EventSource("/extract/stream?file_path=doc.pdf");

eventSource.onmessage = (event) => {
  if (event.data === "[DONE]") {
    eventSource.close();
    console.log("Extraction complete");
  } else {
    document.getElementById("output").innerHTML += event.data;
  }
};

---

Progress Tracking

Track extraction progress for progress bars:

from strutex import DocumentProcessor, GeminiProvider

processor = DocumentProcessor(provider=GeminiProvider())

total_chars = 0
estimated_total = 5000  # Estimate based on document size

for chunk in processor.stream("doc.pdf", "Extract", schema=MySchema):
    total_chars += len(chunk)
    progress = min(total_chars / estimated_total * 100, 99)
    print(f"\rProgress: {progress:.0f}%", end="")

print("\rProgress: 100%")

---

Async Streaming

Use async streaming for better performance:

import asyncio
from strutex import DocumentProcessor, GeminiProvider

processor = DocumentProcessor(provider=GeminiProvider())

async def stream_extract():
    async for chunk in processor.astream(
        file_path="document.pdf",
        prompt="Extract data",
        schema=MySchema
    ):
        print(chunk, end="", flush=True)

asyncio.run(stream_extract())

---

Provider Support

Provider	Streaming Support
GeminiProvider	Full
OpenAIProvider	Full
AnthropicProvider	Full
OllamaProvider	Full
GroqProvider	Limited
ProviderChain	No

---

Best Practices

Practice	Why
Use streaming for docs > 5 pages	Better UX for long operations
Show loading indicator initially	Stream may take 1-2s to start
Handle connection drops	Implement retry on disconnect
Buffer chunks on server	Don't send every character

---

Next Steps

Want to...	Go to...
Handle errors	Error Handling
See use cases	Use Cases
Optimize prompts	Prompt Engineering