YouTube to MP3 with an API — two calls, one signed link

info answers with audio_formats[]: an m4a row with the exact byte size and an mp3 row whose size is an estimate. Title, duration and thumbnail come along, so a bot can show the user what it is about to send before spending 10 credits on the file.

curl -s "https://api.saveapi.org/v1/youtube/info" \
  -H "Authorization: Bearer sk_live_xxxx" \
  -G --data-urlencode "url=https://www.youtube.com/watch?v=dQw4w9WgXcQ"

The same link plus quality=mp3 (or m4a). The response carries url, filename with the title already in it, mode "audio", file_size and expires. Nothing is stored on our side: the bytes are produced as you read the link.

curl -s "https://api.saveapi.org/v1/youtube/create" \
  -H "Authorization: Bearer sk_live_xxxx" \
  -G --data-urlencode "url=https://www.youtube.com/watch?v=aqz-KE-bpKQ" \
  --data-urlencode "quality=mp3"

It is a stream, not a stored file: no Content-Length, no Range, one pass from start to end. After the expires timestamp it answers 410 — call create again rather than keeping links around. Treat it as a download that has to happen now.

Stream the response to disk under the filename you were given, then send the file from your bot or store it where you need it. Send the file, not the link: Telegram fetches URLs itself with its own limits (20 MB for audio by URL), and a stream without a declared size is a poor fit for that.

import httpx

def fetch_audio(made: dict, dest: str) -> str:
    """Pull the file behind a /v1/youtube/create link onto disk."""
    path = f"{dest}/{made['filename']}"
    with httpx.stream("GET", made["url"], timeout=None) as r, open(path, "wb") as f:
        if r.status_code == 410:            # past expires: call create again
            raise RuntimeError("link expired")
        for chunk in r.iter_bytes():        # no Content-Length: read to the end
            f.write(chunk)
    return path

# made = youtube_link(url, "mp3")          # from the yt-dlp guide
# audio = fetch_audio(made, "/tmp")
# bot.send_audio(chat_id, open(audio, "rb"), title=made["title"])

M4A is the AAC track YouTube publishes: exact size, no transcoding, the stream starts immediately. MP3 is built from it with ffmpeg, so it takes longer to begin and its size is estimated. Pick M4A unless the player on the other end insists on MP3 — Telegram plays both.