Deploy Preds in under 10 minutes

Walk through key creation, your first requests and production monitoring.

Prerequisites

Make sure you have the following before sending requests:

Preds workspace with the Free tier or any paid plan.
HTTP client with TLS 1.2+ and JSON support (curl, fetch, HTTPie).

Texts must contain at least 60 words and every call must use HTTPS + JSON.

Base URLs

Use the same version locally and in production.

`https://preds.hu/api/v1` — production, EU-hosted infrastructure.
`Content-Type: application/json` and `Authorization: Api-Key …` are required on every request.

Create an API key

Keys live on a workspace and are only shown once.

Open Dashboard → API keys → “New key”.
Provide a descriptive label (for example `backend-ingest`).
Copy the value into your secrets manager immediately.

# store the Preds key securely
export PREDS_API_KEY="sk_live_..."

Send the first detection

The synchronous endpoint answers in ~200ms and enforces the 60 word minimum.

Python example

import os
import requests

# Configuration
API_KEY = os.getenv('PREDS_API_KEY')
BASE_URL = 'https://preds.hu/api/v1'

# Prepare request
headers = {
    'Authorization': f'Api-Key {API_KEY}',
    'Content-Type': 'application/json'
}

payload = {
    'text': 'Paste the Hungarian text you want to audit (minimum 60 words).'
}

# Send detection request
response = requests.post(
    f'{BASE_URL}/detect/api/',
    headers=headers,
    json=payload
)

# Handle response
if response.status_code == 200:
    data = response.json()
    print(f"Prediction: {data['prediction']}")
    print(f"Confidence: {data['confidence']}")
    print(f"Quota remaining: {data['quota']['remaining']}")
else:
    print(f"Error: {response.status_code} - {response.text}")

Queue async detections

Use the task-backed endpoint for larger batches or workflows where you prefer polling.

Queue a task (Python)

import os
import requests
import time

# Configuration
API_KEY = os.getenv('PREDS_API_KEY')
BASE_URL = 'https://preds.hu/api/v1'

headers = {
    'Authorization': f'Api-Key {API_KEY}',
    'Content-Type': 'application/json'
}

# Queue async detection
response = requests.post(
    f'{BASE_URL}/detect/async/',
    headers=headers,
    json={'text': 'Paste the Hungarian text you want to audit.'}
)

task_data = response.json()
task_id = task_data['task_id']
print(f"Task queued: {task_id}")

# Poll for completion
while True:
    result_response = requests.get(
        f'{BASE_URL}/tasks/{task_id}/result/',
        headers=headers
    )
    result = result_response.json()

    if result['status'] == 'completed':
        data = result['data']
        print(f"Prediction: {data['prediction']}")
        print(f"Confidence: {data['confidence']}")
        break
    elif result['status'] == 'failed':
        print(f"Task failed: {result.get('message')}")
        break

    time.sleep(1)  # Wait before polling again

Poll for results

curl -X GET https://preds.hu/api/v1/tasks/<task_id>/result/   -H "Authorization: Api-Key $PREDS_API_KEY"

{
  "task_id": "b2b6c9c2-1c2e-4e3c-9a3d-43c7c9326d2f",
  "status": "completed",
  "data": { "prediction": "ai", "confidence": 0.91, "submission_id": 1341 }
}

Handle the response

Response format is clean and professional. Use ?detailed=true for full data with segments.

`prediction` and `confidence` form the primary verdict.
`request_id` helps with debugging. In detailed mode (`?detailed=true`), `submission_id` is also available.
docs.gettingStarted.handleResults.items.latency

Hardening checklist

Bake the following into your production deployment.

Rotate API keys quarterly or immediately after an incident.
Use separate keys per service so you can audit scope usage.
Monitor `quota.remaining` and `rate_limit.remaining`.

Troubleshooting

Most integration issues correspond to these signals.

`code: quota_exceeded` → wait for the new cycle or purchase credits, the response includes remaining values.
`code: rate_limit_exceeded` → implement exponential backoff and unique request IDs.
`Text must contain at least 60 words` → surface a friendly error or pad the sample text.
Still blocked? Email [email protected] with the `request_id`.

Next steps

Review the full API reference for async, OCR and reporting endpoints.
Share the contact link if you need help with custom workflows.