Audio inputs

Audio requires blink-1.0-pro. Send one WAV or MP3 data URL in the audio array. Keep the file under 10 MiB; the first 30 seconds are processed. Send only one attachment per request.

Try a customer voicemail

This synthetic voicemail describes a damaged delivery and asks for a replacement. The caller offers to send a photo — that does not establish that support has received it.

Open the example in the console. It selects Pro automatically. Use the Attachment tab to play the recording and the State tab to edit the support policy.

Call the API

Use your existing blnkd API key. This Python example uses HTTP directly, including when your SDK does not expose an audio field.

import base64
import os
from pathlib import Path
import httpx

recording = base64.b64encode(Path("voicemail.wav").read_bytes()).decode()
response = httpx.post(
    "https://api.blnkd.dev/v1/systemone",
    headers={"Authorization": f"Bearer {os.environ['BLNKD_API_KEY']}"},
    json={
        "model": "blink-1.0-pro",
        "state": {"task": "Triage this customer voicemail."},
        "audio": [f"data:audio/wav;base64,{recording}"],
        "questions": {
            "replacement": {
                "type": "noul",
                "instructions": "Does the caller ask for a replacement?"
            }
        },
    },
    timeout=120,
)
response.raise_for_status()
print(response.json())

Use data:audio/mpeg;base64, for MP3. Flash rejects audio requests. Answers are typed decisions, not a transcript.