YouTube Channels

What It Does

YouTube channels brings a creator's videos into a Research Café as sources. You find a channel, tick the videos you want (or let Latest N pick the newest ones for you), press Scrape, and each video's full transcript lands in the café as an ordinary source — something you can read, cite, and ask the café chat about, just like a web page or a paper.

It is for when one video is not enough: a lecture series, a podcast's back catalogue, every interview a creator has given on a topic.

Opening It

There are two ways in, and they open the same screen in the café's centre column:

• ✦ Pro → Gather → YouTube channels on the chat composer.
• ＋ Add sources → YouTube channels ("Turn creators' videos into sources").

Nothing is charged for opening it.

1. Find a Channel

Type into the one box at the top. It understands three things:

• An @handle — for example @veritasium.
• A channel link — youtube.com/@veritasium, youtube.com/channel/UC…, or an older /c/… or /user/… link.
• A creator's name — for example Veritasium. A name is a search (it can take up to a minute — the screen shows each step and a seconds counter while it works): you get a short list of Matching channels, ranked by popularity, each with an Add button, and you add the one you meant. Once you add one, the list closes and the box clears.

A handle or a channel link goes straight to that channel with no search. A link to a single video is not a channel; paste a channel's @handle or link instead (or use the café search bar, which extracts one video's transcript).

You can add several channels and pick from all of them in one go. The channel you just added appears at the top, open; the channels you added earlier fold down to a one-line header showing how many videos are listed and how many you've ticked — click a header to open it again. Folding a channel never loses your ticks. Each channel has an ✕ to remove it.

2. Pick Videos

Each channel shows its newest 15 regular videos (Shorts and live streams are not listed). Load 15 more goes further back, up to 300 videos per channel. Every row shows the thumbnail, title, how long ago it was posted, its length, its views, and an estimate of the tokens its transcript will cost.

• In café — a video that is already a source in this café is greyed out and can't be ticked, so you never pay twice for the same transcript.
• Long — a video over an hour carries a Long badge. Long videos are kept whole (the full transcript, not a cut-down one), so they take longer to index; the footer warns you when your pick includes any.
• Quick pick: Latest 5 / 10 / 15 — ticks that channel's newest videos that are not already in the café. If the shown rows don't hold enough, it loads the next pages for you first. It replaces only that channel's ticks, so you can mix a hand-picked selection from one channel with "Latest 10" from another.

The footer shows how many videos you've picked, across how many channels, and the estimated tokens, before you commit.

Listing is rate-limited to protect your balance: if you press Load 15 more (or Latest N pulls pages) very rapidly — more than 30 listings in 10 minutes — you'll see "You're loading channel videos very quickly. Please wait a few minutes, then try again." Nothing is charged for a refused request, and a normal walk through a channel never reaches the limit.

3. Scrape

Press Scrape N videos. The scrape runs in the background — you can leave the café, open something else, or close the tab, and come back. When you reopen the screen, it picks up the running scrape's progress where it is.

While it runs you see "X of N done · T tokens so far" and one row per video:

• ✓ Added — the transcript is now a café source, with the tokens it cost.
• ✕ No transcript — not charged — the creator turned captions off, so there is nothing to add. You pay nothing for it.
• Video not found — not charged — the video was removed or made private.
• Already in café — not charged — the video became a source here between picking and scraping (for example from another tab), so it is skipped.
• Out of tokens — not charged — your balance ran out partway. These videos were never fetched or charged, and they are retryable: top up, then press Retry failed.
• Refund pending — retried automatically / Refund pending — not charged again — see "One Scrape at a Time" below.
• Failed — a temporary problem fetching it. Only this plain label reads red; the informational rows above do not.
• ⟳ fetching… and · waiting — still to come. Up to four videos are fetched at once.

Your café's source list fills in as videos are added — you don't have to wait for the whole scrape to finish to start reading or asking about them.

Cancel stops the scrape: videos already being fetched finish (and are added), nothing new starts, and the videos that never started are marked "Cancelled — not charged". If anything Failed (a temporary problem fetching it) or ran Out of tokens, Retry failed starts a new scrape over just those videos — never the ones with no transcript, which would only give the same answer again. Back to channels returns you to your pick list.

In This Café: What You've Already Scraped

Under the Find channel box, In this café lists every creator whose videos are already in this café — newest first, each with its picture (or its first letter), its name, and "N videos · last scraped when". Click a creator to see their videos (thumbnail, title, and when it was added); click a video to open it in the café.

If a scrape finished while you were away — you closed the tab, or opened something else — a Last scrape line at the top of the list tells you how it went, for example "Last scrape · 2m ago — 4 added, 1 had no transcript (not charged)". Failed or cancelled videos are counted there too. Click ✕ to dismiss it. Nothing is ever lost by closing the tab: every video that finished is in the café.

When you find or add a creator, their videos open on top and the list folds to a single line ("In this café · 3 creators") underneath — click it to open the list again.

One Scrape at a Time

You can run one scrape at a time across your whole account. If you press Scrape while one is still running:

• In the same café, the screen simply shows the running scrape.
• In a different café, you see that scrape's progress read-only, headed "Scraping in café name", with a Go to café name link to manage it there. Your ticks in this café are kept, and Scrape is available again as soon as the other scrape finishes.

A scrape that stops making progress (for example, if our server restarts mid-run) is closed out automatically after a while, so it never blocks you: every video that finished was added and charged once, any video cut off mid-way is refunded automatically, and Retry failed picks up the rest.

If a refund ever can't be made straight away (a temporary problem on our side), that video shows Refund pending — retried automatically with its charge still visible. CoffeeScribe keeps retrying the refund each time you open the scrape or start a new one, and the video is never charged a second time: a new scrape that includes it marks it Refund pending — not charged again and skips it until the refund has gone through.

What It Costs

You pay in tokens, and only for what you get:

• Searching by name and listing a channel's videos cost a small amount per page of results (each page of 15 is charged once — Load 15 more charges only for the new rows). A handle or a link skips the search.
• Each transcript is charged once, and kept only when it is added to the café. As each video starts, its charge is set aside first; if the video turns out to have no transcript, isn't found, fails, is cancelled, or can't be saved, that charge is refunded automatically (safely — a refund can never credit you twice). So "No transcript", "Video not found", cancelled and failed videos cost you nothing (if one shows "Refund pending", its refund is on its way automatically).
• Before a scrape starts, CoffeeScribe checks your balance covers every video you picked (each transcript is charged on its own), plus each channel's listing; if it doesn't, the scrape doesn't start (and nothing is charged). If your balance is spent elsewhere partway through, the remaining videos end Out of tokens — not charged and can be retried after a top-up. If you use your own Apify key, this check is skipped.

Using your own Apify key (BYOK)

If you've added your own Apify key in Settings → API Keys, the listing and the transcripts run on your key and are not charged in tokens — you pay Apify directly. If your key stops working mid-scrape, CoffeeScribe finishes on its own key and charges tokens from that point, so the scrape still completes. See Using your own API keys (BYOK).

Before You Start: Social Sources

YouTube is a social platform source, so this screen needs your account's Social platform sources switch to be on. If it's off when you find a channel, load videos or press Scrape, the Social Sources Terms dialog opens right there. Accept and the action you were doing carries on automatically; decline and nothing is searched, listed or charged. You can also change it any time in Settings → Data & Privacy → Social platform sources. See Research Mode for how that switch works.

Asking About What You Scraped

Each added video is a normal café source titled "video title — channel", linking to the video. In the café's Sources list, your videos sit under YouTube, in one folder per channel — for example Veritasium · 4. Open a folder to see that channel's videos; inside it each title drops the "— channel" ending and wraps onto two lines instead of being cut off (hover a title to see it in full). Videos whose channel isn't known — for example ones you added by pasting a single link — are under Other videos. Ask the café chat a question and it answers from the transcripts like any other source, citing the video it drew on.