tinyfish

TinyFish Web Agent

Requires: MINO_API_KEY environment variable

Best Practices

1. Specify JSON format: Always describe the exact structure you want returned
2. Parallel calls: When extracting from multiple independent sites, make separate parallel calls instead of combining into one prompt

Basic Extract/Scrape

Extract data from a page. Specify the JSON structure you want:

import requests
import json
import os

response = requests.post(
    "https://mino.ai/v1/automation/run-sse",
    headers={
        "X-API-Key": os.environ["MINO_API_KEY"],
        "Content-Type": "application/json",
    },
    json={
        "url": "https://example.com",
        "goal": "Extract product info as JSON: {\"name\": str, \"price\": str, \"in_stock\": bool}",
    },
    stream=True,
)

for line in response.iter_lines():
    if line:
        line_str = line.decode("utf-8")
        if line_str.startswith("data: "):
            event = json.loads(line_str[6:])
            if event.get("type") == "COMPLETE" and event.get("status") == "COMPLETED":
                print(json.dumps(event["resultJson"], indent=2))

Multiple Items

Extract lists of data with explicit structure:

json={
    "url": "https://example.com/products",
    "goal": "Extract all products as JSON array: [{\"name\": str, \"price\": str, \"url\": str}]",
}

Stealth Mode

For bot-protected sites:

json={
    "url": "https://protected-site.com",
    "goal": "Extract product data as JSON: {\"name\": str, \"price\": str, \"description\": str}",
    "browser_profile": "stealth",
}

Proxy

Route through specific country:

json={
    "url": "https://geo-restricted-site.com",
    "goal": "Extract pricing data as JSON: {\"item\": str, \"price\": str, \"currency\": str}",
    "browser_profile": "stealth",
    "proxy_config": {
        "enabled": True,
        "country_code": "US",
    },
}

Output

Results are in event["resultJson"] when event["type"] == "COMPLETE"

Parallel Extraction

When extracting from multiple independent sources, make separate parallel API calls instead of combining into one prompt:

Good - Parallel calls:

# Compare pizza prices - run these simultaneously
call_1 = extract("https://pizzahut.com", "Extract pizza prices as JSON: [{\"name\": str, \"price\": str}]")
call_2 = extract("https://dominos.com", "Extract pizza prices as JSON: [{\"name\": str, \"price\": str}]")