Get a Daily YouTube Digest From Your AI Agent Instead of Missing Videos
On this page
YouTube knows you subscribed to those channels. It just doesn’t care. The algorithm’s job isn’t to show you what you asked for — it’s to show you what keeps you on the platform longest. Your favorite educator who posts twice a week? Buried. A trending reaction video from someone you’ve never heard of? Front and center.
I’ve watched this pattern frustrate people for years. They subscribe to thoughtful channels — technical deep-dives, industry analysis, specific niche content they actually chose — and then those channels just vanish from their feed. Not because the creators stopped posting. Because YouTube decided something else was more engaging.
The fix isn’t checking YouTube more often. It’s stopping the manual process entirely and letting software handle it. If you’re curious about what AI automation can actually do for your daily information intake, this is one of the more practical examples I’ve seen. I’ll get to the part that genuinely surprised me about how cheap this is to run — after I walk through the setup.
Why YouTube’s Subscription Feed Is Working Against You
YouTube mixes algorithmic recommendations with your subscriptions. The feed isn’t your subscription list — it’s a blend of what you asked for and what YouTube thinks will keep you watching. That’s the core problem.
The numbers on this are rough but telling. An average professional subscribes to 15–30 educational YouTube channels. Each channel posts 2–4 videos a week. That’s 30–120 new videos weekly — 6–24 hours of content, or 3–12 hours even at double playback speed. You’re not going to watch all of that. The real question is how you decide what’s worth your time.
Research on the topic frames the core problem as decision fatigue, not lack of time. You waste mental energy figuring out what to watch, which videos matter, and what you can skip. By the time you’ve decided, you’re already tired — and that’s before you’ve hit play on anything.
A daily digest solves this differently. Instead of you deciding in real-time while scrolling a feed designed to distract you, software checks for new uploads overnight, pulls the key points from each video’s text track, and sends you a summary. You read 60 seconds of bullets over coffee and decide what’s actually worth watching — if anything.
What an AI Digest Actually Does
The workflow is straightforward. Your agent — running on a schedule you set — does three things: checks your list of channels for new uploads, grabs the text version of each video (the transcript), and summarizes the key points into a few bullets.
The result hits your phone or inbox every morning. Video title, channel name, link, and 2–3 bullet points on what was covered. You’re not watching the video — you’re reading a briefing. If it sounds relevant, you watch. If not, you’ve saved 20 minutes.
This runs on OpenClaw — the agent framework that powers BrainRoad’s AI automation platform. The specific skill you need is called youtube-full, available from ClawHub. It handles transcript fetching, channel resolution, and delivery — your agent just needs instructions on what channels to watch and when to run.
How to Set Up Your Daily YouTube Digest
First, install the skill. Tell your OpenClaw agent:
Install the youtube-full skill and set it up for me
Or run it directly from your terminal:
npx clawhub@latest install youtube-full
You get 100 free credits on signup, no credit card required. The skill auto-stores your API key in the right location for your operating system — it works the same whether your agent is running locally or in the cloud. Now pick your setup:
Beacon says: your favorite creators never go on without you — just tell your agent what to watch for, and let the light do the rest.
Option 1: Channel-Based Digest (Most People Start Here)
Give your agent this prompt — swap in the channels you actually follow:
Every morning at 8am, fetch the latest videos from these YouTube channels and give me a digest with key insights from each:
- @TED
- @Fireship
- @ThePrimeTimeagen
- @lexfridman
For each new video (uploaded in the last 24-48 hours):
1. Get the transcript
2. Summarize the main points in 2-3 bullets
3. Include the video title, channel name, and link
If a channel handle doesn't resolve, search for it and find the correct one.
Save my channel list to memory so I can add/remove channels later.
Your agent saves the channel list to memory. You can add or remove channels later by just telling it. No config files, no YAML.
Option 2: Keyword-Based Digest (For Tracking Topics, Not Just Channels)
If you’re tracking a topic — “AI agents”, “Rust programming”, “options trading” — rather than specific creators, use this pattern:
Every day, search YouTube for new videos about "OpenClaw" (or "Claude Code", "AI agents", etc).
Maintain a file called seen-videos.txt with video IDs you've already processed.
Only fetch transcripts for videos NOT in that file.
After processing, add the video ID to seen-videos.txt.
For each new video:
1. Get the transcript
2. Give me a 3-bullet summary
3. Note anything relevant to my work
Run this every morning at 9am.
The Cost Breakdown That Actually Surprised Me
Here’s the thing most people don’t realize until they look at the API docs: checking for new uploads is free. The channel/latest and channel/resolve endpoints cost zero credits. Your agent can scan every channel on your list every morning without spending anything.
You only pay when you pull a transcript — and that costs 1 credit per video. If you follow 10 channels and each posts once a week, you’re spending 10 credits per week on transcripts. With 100 free credits on signup, your first two months of the channel-based digest are covered before you pay anything.
For the keyword-based digest, the seen-videos.txt file is what keeps costs sane. Without it, you’d refetch transcripts for videos that haven’t changed. With it, the agent skips anything it’s already processed. This is the kind of detail that separates a setup that costs pennies from one that quietly runs up your credit balance.
Why yt-dlp Breaks and the youtube-full Skill Doesn’t
If you’ve tried to build something like this before, you probably started with yt-dlp. It seems obvious — it’s free, it’s widely used, and it handles YouTube downloads. The problem is what happens when you run it inside a cloud agent.
yt-dlp dumps verbose logs into your agent’s active memory. Every status message, every progress update — it all gets fed into the context your agent is working from. On a long channel list, this fills up the working memory fast and degrades output quality. That’s the best-case scenario.
The worse case: yt-dlp gets blocked. YouTube actively limits CLI tools, and if you’re running an agent on cloud infrastructure (Google Cloud, AWS, any managed platform), the requests often fail silently or return errors. It’s not a bug in your setup — it’s YouTube actively filtering that traffic.
The youtube-full skill uses TranscriptAPI, which returns clean JSON — no log noise, no context pollution. It works from cloud infrastructure because it’s built for that context, and it’s cached and reliable at scale. The tool that powers YouTubeToTranscript.com handling millions of requests is a different category of reliability than a CLI tool designed for personal downloads.
How You Know It’s Working
- You receive a digest message at your scheduled time with video titles, channel names, and bullet summaries
- Channels that haven’t posted in the last 24–48 hours produce no transcript charges — only the free channel-check runs
- For keyword digests, seen-videos.txt grows each day as new video IDs get appended
- If you ask your agent to add or remove a channel, it confirms the update and the next morning’s digest reflects the change
- You can ask for the digest on-demand any time, not just at the scheduled hour — the agent runs the same workflow immediately
Where This Breaks Down
- Channels with auto-generated captions only — transcripts exist but quality varies. Short or heavily accented content sometimes produces summaries that miss context.
- Videos under ~5 minutes often don’t have enough transcript content for meaningful bullets. Your agent will still try, but the output may be thin.
- If a channel changes its handle, the agent may fail to resolve it until you manually tell it the new handle.
- Keyword-based digests on broad terms (“AI”, “programming”) will return too many videos to summarize meaningfully. Narrow your keywords.
- The 100 free credits go fast if you add 20+ channels on day one. Start with 5–10 channels you actually care about, then expand.
Your Monday Morning Digest Setup Checklist
- Install the youtube-full skill using
npx clawhub@latest install youtube-full— takes under 2 minutes. You’ll get 100 free credits automatically with no credit card required. - Pick 5–10 channels you actually miss content from. Don’t start with 30 — you’ll overwhelm the digest and burn credits fast.
- Choose your delivery time. 8am works for most people. If you’re in a different timezone, calculate the UTC offset and tell your agent explicitly (e.g., ‘8am Eastern time’).
- If you’re tracking a topic instead of specific channels, add the seen-videos.txt instruction from Option 2 above on day one — not after you’ve already burned credits on repeats.
- If your credit balance drops faster than expected, check whether any channel on your list posts multiple videos daily. One channel posting 5 videos a day will cost 35 credits a week on its own.
- After 48 hours, ask your agent: ‘Show me the channels that produced the most transcript credits this week.’ Use that to decide if any channels aren’t worth the credit cost.
- For more on building broader information pipelines with your agent, check out Build an AI Content Pipeline That Works While You Sleep.
What This Means for Your Information Diet
- YouTube’s algorithm actively buries content from your subscribed channels — the missing videos aren’t a glitch, they’re by design.
- An AI digest agent can check your channel list for new uploads every morning at zero cost — transcript pulls cost 1 credit each, and 100 free credits come with signup.
- The seen-videos.txt deduplication pattern is critical for keyword-based digests — without it, you’ll pay to re-summarize the same videos daily.
- CLI tools like yt-dlp fail on cloud infrastructure and flood agent memory with logs — the youtube-full skill on ClawHub solves both problems with clean JSON responses.
- Start with 5–10 channels and one scheduled run per day. The setup takes under 10 minutes and works with the OpenClaw agents that power BrainRoad’s personal AI assistant platform.
Frequently Asked Questions
Do I need a YouTube account or API key to run this?
No YouTube account or YouTube API key is required. The youtube-full skill uses TranscriptAPI, which has its own account system. You create a free TranscriptAPI account during setup and get 100 credits automatically. The skill handles the API key storage — you don’t touch any config files.
What if a video doesn't have a transcript?
Most YouTube videos have auto-generated captions, which the skill can pull as a transcript. If a video has captions disabled or is in a language without auto-generation, the transcript fetch will fail gracefully — your agent will note that video and move on. You won’t be charged credits for failed fetches.
Can I get the digest on WhatsApp or Signal instead of email?
Yes. If your OpenClaw agent is connected to a messaging channel — WhatsApp, Signal, Telegram, or others — you can have the digest delivered there instead of or in addition to email. Just tell your agent where to send it when you set up the scheduled prompt. BrainRoad agents support multiple delivery channels out of the box.
How is this different from services like SumyTube or Recapio?
Dedicated services like SumyTube let you track up to 5 channels and send summaries via email. They’re simpler to set up if you just want email delivery. The OpenClaw approach gives you more control: unlimited channels, keyword tracking, custom delivery channels, scheduled timing, and the ability to extend the workflow to other tasks. If you want a zero-configuration option and 5 channels is enough, a dedicated service works. If you want the digest to fit into a broader personal AI agent setup, the OpenClaw route is more flexible.
Will this work if my agent is running in the cloud rather than locally?
Yes — and this is specifically why the youtube-full skill uses TranscriptAPI instead of yt-dlp. CLI tools like yt-dlp get blocked by YouTube when requests come from cloud IP ranges (Google Cloud, AWS, etc.). TranscriptAPI handles cloud deployments without issues. The skill auto-stores your API key in the correct location for both local and cloud environments.
