Access guide
Slow yt-dlp down to avoid 429s and bot checks
Hammering YouTube with back-to-back requests is one of the fastest ways to earn an HTTP 429 Too Many Requests, a "Sign in to confirm you're not a bot", or a "you must log in" wall. Pacing your requests makes you look less like a scraper and buys you a lot of headroom. Here are the flags that do it, values that actually work, and the one thing pacing can't fix.
Put time between requests
The two flags that matter most are --sleep-requests (a pause between the metadata/API requests yt-dlp makes) and --sleep-interval / --max-sleep-interval(a random pause before each download). The random range is what matters — a fixed, identical gap is itself a fingerprint.
# ~0.75s between requests, plus a random 10-20s pause before each download
yt-dlp \
--sleep-requests 0.75 \
--sleep-interval 10 \
--max-sleep-interval 20 \
"URL"With --sleep-requests set, yt-dlp prints a Sleeping N seconds ... line between the player/API calls, so you can see it working. On a whole playlist or channel, these pauses are the single biggest lever against a block.
Cap the bandwidth and the parallelism
A flat-out multi-connection download also stands out. Cap the rate with --limit-rate and pull one fragment at a time with --concurrent-fragments 1 (the default is already 1, so this really means: don't crank it up):
yt-dlp \
--limit-rate 2M \
--concurrent-fragments 1 \
--sleep-requests 0.75 \
"URL"One operator on a large channel reported that throttling to well under a megabyte a second kept their IP off the "you must log in" wall — the job ran slower but finished, where going flat-out got them blocked. Slow and complete beats fast and banned.
Back off on retries, don't pound
When something does fail, retrying immediately makes it worse. --retry-sleep adds a growing gap between attempts, so a transient 429 gets a chance to clear instead of hardening into a block:
# linear backoff on HTTP errors, exponential on fragment retries
yt-dlp \
--retry-sleep linear=1::2 \
--retry-sleep fragment:exp=1:20 \
"URL"Verified against yt-dlp 2026.10.26. If you are already seeing 429s, also make sure you are on a current build — see updating yt-dlp.
What pacing can't fix
Pacing lowers how often you trigger a check. It does nothing about where you're coming from. If you are on a server, CI runner, or agent sandbox, the IP is a datacenter IP, and datacenter ranges are flagged by reputation before you send a single request — so you can hit a bot check or "Sign in to confirm" on the very first call, no matter how gently you pace it. That's a different problem:
- yt-dlp works locally but not on my server — why the same command fails once it leaves your laptop.
- using a proxy with yt-dlp — residential vs datacenter, the distinction that actually decides it.
- fixing "Sign in to confirm you're not a bot" — cookies and the rest of the ladder.
The honest free fix first
On your own machine, pacing plus an up-to-date yt-dlp (and cookies when a video needs them) is usually enough — start there, it costs nothing. The paid problem only shows up when you have to run on a server at volume, where a flagged IP means pacing alone will never clear the wall and you're into residential proxies and rotation to maintain. If the job only needs the file, metadata, or transcript rather than yt-dlp itself, a hosted API that already runs on rotating residential IPs takes the IP and pacing problem off your side of the line. That's the honest case for Importly for server and agent workloads.
Servers & agents
Works on your laptop but not on the server?
Paste a link to see what a hosted API returns from a clean residential IP — no cookies, proxies, or runtime to maintain on the box.