DEV Community: Ahror

[Boost]

Ahror — Wed, 14 Oct 2026 15:51:55 +0000

Ahror

Oct 14

Why your Instagram scraper works on your laptop and dies on the server

#webscraping #python #devops #instagram

3 min read

How to build a Telegram video downloader bot

Ahror — Wed, 14 Oct 2026 15:25:39 +0000

A link-in, video-out bot is about forty lines once the resolving is somebody else's problem. This guide is the whole path: link to media URL, media URL to a Telegram message, and the three things that break in production — private posts, files Telegram refuses to fetch, and re-sending the same video a hundred times.

1. Install the two libraries

aiogram talks to Telegram, httpx makes the API call. Nothing else is needed on your side: there is no scraping stack, no headless browser and no cookie jar to keep alive.

pip install aiogram httpx

2. Resolve the link

One GET with your key in the Authorization header. The endpoint detects the platform from the URL itself, so the bot does not branch on the domain — an Instagram reel, a TikTok video and a Likee clip all go through this same function.

import httpx

API = "https://api.saveapi.org/v1/download"
KEY = "sk_live_xxxx"

client = httpx.AsyncClient(timeout=30)

async def resolve(link: str) -> dict:
    r = await client.get(
        API,
        params={"url": link},
        headers={"Authorization": f"Bearer {KEY}"},
    )
    return r.json()

3. Answer with the media

The response tells you what it is. Check medias[0].type instead of assuming a video: a TikTok slideshow and an Instagram carousel come back as images, and a Telegram bot has to call a different method for those. meta.title makes a reasonable caption.

from aiogram import Bot, Dispatcher, F
from aiogram.types import Message

dp = Dispatcher()

@dp.message(F.text.startswith("http"))
async def on_link(message: Message):
    data = await resolve(message.text.strip())

    if not data["success"]:
        await message.answer(EXPLAIN.get(data["error"]["code"], "Try another link."))
        return

    media = data["medias"][0]
    if media["type"] == "video":
        await message.answer_video(media["url"], caption=data["meta"]["title"])
    else:
        await message.answer_photo(media["url"])

4. Explain the errors people actually hit

Users paste private posts and dead links all day. Map the stable error codes to sentences a person understands — a bot that says "That account is private" gets a different link, a bot that says "Error 403" gets an angry message. A private post or one with no media costs 0.1 credits instead of the full 1.5.

EXPLAIN = {
    "PRIVATE_CONTENT": "That account is private — I can only fetch public posts.",
    "MEDIA_NOT_FOUND": "That post has no media, or it was deleted.",
    "UNSUPPORTED_PLATFORM": "I do not know that site yet.",
    "QUOTA_EXCEEDED": "Out of credits for today.",
    "UPSTREAM_TIMEOUT": "The source was slow. Send the link again.",
}

5. Cache Telegram's file_id, not the media URL

Once Telegram has the file it gives you a file_id, and re-sending that costs nothing and uploads nothing. The media URL is the wrong thing to store: source CDNs sign their links and expire them within hours, so a cache of URLs turns into a cache of 403s overnight.

sent = await message.answer_video(media["url"])
file_id = sent.video.file_id      # BU saqlanadi, media["url"] emas

# Keyingi safar o'sha post so'ralsa:
await message.answer_video(file_id)

6. Fall back to uploading bytes

Telegram fetches the URL itself, and sometimes the source CDN refuses it — most often for Instagram, and always in a media group. Catch that case, download the bytes yourself and upload them. The 50 MB bot upload limit still applies, which is why this is a fallback and not the default path.

from aiogram.types import BufferedInputFile

async def send_bytes(message: Message, url: str):
    r = await client.get(url, follow_redirects=True)
    file = BufferedInputFile(r.content, filename="video.mp4")
    await message.answer_video(file)

FAQ

Can the bot send files larger than 50 MB?

Not by uploading them: 50 MB is Telegram's ceiling for a bot upload. The usual way around it is to send the media URL and let Telegram fetch it, or to keep large files in a private channel and copy the message by file_id.

Do I need a separate handler for each platform?

No. /v1/download detects the platform from the link, so one handler covers every supported site. Per-platform endpoints exist for when you deliberately want to accept only one.

What happens when a user pastes a private post?

You get a 403 with the PRIVATE_CONTENT code and it costs 0.1 credits instead of the full 1.5. Turn it into a sentence: people paste private links constantly and they are not doing anything wrong.

This guide was first published on saveapi.org. The API used here has a free tier — 1,000 credits, no card: get a key. Docs: saveapi.org/en/docs.

Prefer a client library? pip install saveapi (Python) or npm install saveapi (JavaScript/TypeScript) — github.com/akhroruz/saveapi-sdk.

Why your Instagram scraper works on your laptop and dies on the server

Ahror — Wed, 14 Oct 2026 15:24:48 +0000

The most common message in our support inbox is a variation of “it worked yesterday on my machine”. It did, and it will again — on that machine. Instagram, TikTok and the sites that front them treat a Hetzner or AWS address differently from a home connection, and a scraper that passed every local test meets that difference in production. This guide is what the difference is, measured, and how to build around it rather than against it.

1. The address is the fingerprint

Platforms score a request before reading it: a datacenter ASN, a TLS handshake that does not match the claimed browser, and a session that does not exist all lower the score, and the page you get is the login wall or a captcha. Your laptop passes on all three without trying. The server fails the first one before your code runs.

2. What we measured

Instagram's own endpoints from a Hetzner IP: 0 of 11 posts. yt-dlp: 0 of 13. The download sites that serve as intermediaries went the same way one at a time — one family of them put a Turnstile captcha in front of datacenter addresses and dropped from 11 of 11 in July to 0 of 8 in September 2026. Every one of those links resolved from a home connection the same day.

3. Why a proxy is a treadmill, not a fix

A residential proxy restores the score for a while and then the pool gets flagged, because everyone else is renting the same addresses. TLS fingerprinting needs a client that impersonates a browser handshake — plain requests or httpx is refused with a 422 before the body is read. Each fix is another moving part to watch, and the platform changes faster than you patch.

4. What a resolver that lives with it looks like

Ours runs several independent sources in a chain, measures each one from the server address it actually runs on, re-orders them when a source degrades and drops the ones that stop working. You call one endpoint and the chain is our problem. The call from your side is the same one for every platform.

curl -s "https://api.saveapi.org/v1/instagram" \
  -H "Authorization: Bearer sk_live_xxxx" \
  -G --data-urlencode "url=https://www.instagram.com/reel/ABC123/"

5. Handle the two failures that remain

No chain makes a deleted post exist. The response distinguishes: a link the source refuses comes back with a final code and costs 0.1 credits; a timeout or an upstream error costs nothing and is worth exactly one retry. Keep the retryable codes in one place and never retry a final answer.

RETRY = {"UPSTREAM_TIMEOUT", "UPSTREAM_ERROR"}

async def resolve(link: str) -> dict:
    for attempt in (1, 2):
        r = await client.get(API, params={"url": link}, headers=HEADERS)
        data = r.json()
        if data["success"] or data["error"]["code"] not in RETRY:
            return data
    return data

FAQ

Will a VPN on the server help?

Consumer VPN exits are datacenter addresses too, and shared ones are flagged faster than yours. The measurement that matters is the one from the address you deploy on.

Does the API run on datacenter IPs?

Yes — the same kind you deploy on. The difference is that the sources it uses are chosen and measured for that address, and replaced when they stop working, which is the work you would otherwise be doing.

Which platforms are affected?

Instagram is the extreme case, and TikTok's own endpoints behave the same way; sites that publish open JSON, like Pinterest, are largely indifferent. The front page shows the measured success rate across all of them.

This guide was first published on saveapi.org. The API used here has a free tier — 1,000 credits, no card: get a key. Docs: saveapi.org/en/docs.

Prefer a client library? pip install saveapi (Python) or npm install saveapi (JavaScript/TypeScript) — github.com/akhroruz/saveapi-sdk.