Reel Knowledge Distiller

Turn public short videos into transcripts, typed claims, and Markdown notes for research and content teams.

What can Reel Knowledge Distiller do?

Add public YouTube, TikTok, or Instagram video links. The Actor fetches each video, transcribes its speech, sorts its claims into your categories, and saves both dataset rows and Markdown pages.

What you get	Features
📝 A transcript and typed claims for each video	🎬 Direct video and profile input for supported public sources
📚 Category pages and an index in Markdown	🔊 Local Whisper transcription by default

Who this is for

• Build a source set from short videos
• Track claims made by creators or competitors
• Sort video research into topic pages

What you get back

Field	Type	What you get	Example
shortcode	string	Short identifier taken from the source video URL.	fwBIZRq-vzY
url	string	Public video page analyzed by the Actor.	https://www.youtube.com/watch?v=fwBIZRq-vzY
author	string	Creator handle reported for the video.	@Fireship
author_name	string	Creator name when the source provides one.	Fireship
posted	string	Date the video was published when available.	2024-09-21
status	string	Whether this item completed its requested analysis.	distilled
title	string	Title reported for the source item.	Linux command tricks for faster coding workflows
theme	string	Main subject assigned to the video.	linux productivity
categories	array	Knowledge categories assigned to the result.	["ai-production"]
claims	array	Structured claims extracted from the transcript.	[{"type":"technique","signal":"actionable","status":"definitive","text":"Use mkdir { } to create a directory and immediately cd into it, then cd - to return...
tools_mentioned	array	Products or tools named in the source.	["Atom One Dark","vscode-icons","Fira Code"]
why_it_matters	string	Why the extracted material may matter to the selected audience.	These Linux shortcuts streamline directory navigation, bulk file creation, log monitoring, and command reuse , directly boosting developer efficiency.
org_relevance	string	How the result relates to the chosen perspective.	The techniques illustrate practical automation patterns relevant to AI production workflows that the research team can benchmark or integrate into tooling.
importance_signal	string	Signal used to judge why the source may be useful.	The creator highlights overlooked command-line efficiencies, signaling that practitioners value quick, tangible productivity gains.
quality	integer	Quality score assigned to the distilled result.	5
transcript	string	Full speech-to-text transcript.	This cooking trick recently changed my life. If you just drop an egg to crack it, instead of tapping it while holding it, you'll never get shells in the resu...
transcript_source	string	Method used to obtain the transcript.	local-whisper
transcript_chars	integer	Character count of the transcript.	976
caption	string	Source caption or post text when available.	Here are five Linux command tricks that will change your life if you're a programmer. mkdir { } cd - touch {1..100} tail -f history 🎨 My Editor Settings - At...
error	null	Problem details when this item does not complete.	null
error_stage	null	Processing step that failed, if any.	null
model	string	Model used for the analysis.	nvidia.nemotron-nano-3-30b
distilled_at	string	Time the distilled result was created.	2026-09-11T17:11:28Z

The run also links to its dataset and any files named in the Actor output.

What you need to provide

Field	Type	Required	What it does	Example
reelUrls	array	No	Public short-video links we fetch ourselves with yt-dlp , Instagram reels/posts (share → copy link), TikTok videos, YouTube Shorts. No third-party scrapers, no login. Note: Instagram currently serves media only intermittently to non-logged-in clients; failures are reported per row as error_stage: fetch, never faked.	["https://www.youtube.com/shorts/fwBIZRq-vzY"]
reelUrlsFile	string	No	Public URL of a JSON, CSV or text file containing reel/post links (for example an export from your own tooling). Every Instagram/TikTok/YouTube link found in it is added to reelUrls.	``
profiles	array	No	Public YouTube channel URLs (e.g. https://www.youtube.com/@Fireship → its Shorts tab) or TikTok profile URLs (https://www.tiktok.com/@handle). Up to resultsLimit latest videos are listed and processed. Instagram handles are DEPRECATED and skipped with a warning: listing an Instagram profile needs a login, and this Actor never calls third-party scrapers , paste the reel links into reelUrls instead.	``
resultsLimit	integer	No	Cap on how many URLs are processed in one run (1-500). Remaining URLs can be run later; already-processed shortcodes are skipped when you reuse the dataset.	30
perspective	string	No	Free text used in the prompt for org_relevance, e.g. "an AI infra startup's GTM team" or "a solo indie-hacker building a SaaS".	a research team tracking practitioner knowledge
categories	array	No	Knowledge categories the LLM assigns (1-3 per reel, primary first). Each item: {"id": "gtm-sales", "description": "..."}. Leave empty for the built-in default set (ai-production, ai-architecture, model-research, market-state, gtm-sales, marketing-content, product-design, founder-ops, psychology-research, other). An other category is always appended.	``
llm	object	No	Distillation model. Default (when left blank): the Actor's own inexpensive Bedrock model (nvidia.nemotron-nano-3-30b) , no key needed. Override with any /v1/chat/completions server (provider openai	openai-compatible, baseUrl, model, apiKey stored encrypted).
transcription	object	No	mode: local (default , faster-whisper runs inside the Actor on CPU, no key; optional model: tiny	base
minTranscriptChars	integer	No	Transcripts shorter than this are treated as no speech (music/text-overlay reels) and are NOT sent to the LLM.	80

Quick start

1. Open the Actor in Apify Console.
2. Click Try for free or Create a task.
3. Replace the sample values with your own input.
4. Click Start.
5. Open the dataset and the named output files when the run ends.

Pricing

• reel-distilled: $0.01 per reel distilled.
• reel-transcribed: $0.006 per reel transcribed.
• Failed or skipped work is not charged unless an event is listed in the run charges.
• Normal Apify compute and proxy costs may also apply.

Limits and honest notes

• Public platforms can block or rate limit downloads. Each blocked item gets an error row.
• Instagram profile listing needs a login, so the Actor skips Instagram handles. Paste public reel links instead.
• Videos with transcripts shorter than minTranscriptChars return no_speech and are not sent to the LLM.

Code and API

The examples below use the same values as the Apify Console sample.

Input JSON

{
  "reelUrls": [
    "https://www.youtube.com/shorts/fwBIZRq-vzY"
  ],
  "resultsLimit": 30,
  "perspective": "a research team tracking practitioner knowledge",
  "llm": {
    "provider": "openai-compatible",
    "model": "nvidia.nemotron-nano-3-30b",
    "baseUrl": "https://bedrock-runtime.us-east-1.amazonaws.com/openai/v1",
    "apiKey": ""
  },
  "transcription": {
    "mode": "local"
  },
  "minTranscriptChars": 80
}

Real output sample

This excerpt comes from the real run named in the current marketplace release report.

{
  "shortcode": "fwBIZRq-vzY",
  "url": "https://www.youtube.com/watch?v=fwBIZRq-vzY",
  "author": "@Fireship",
  "author_name": "Fireship",
  "posted": "2024-09-21",
  "status": "distilled",
  "title": "Linux command tricks for faster coding workflows",
  "theme": "linux productivity",
  "categories": [
    "ai-production"
  ],
  "claims": [
    {
      "type": "technique",
      "signal": "actionable",
      "status": "definitive",
      "text": "Use mkdir { } to create a directory and immediately cd into it, then cd - to return to the previous directory."
    },
    {
      "type": "technique",
      "signal": "actionable",
      "status": "definitive",
      "text": "Create multiple files with touch {1..100} to generate 100 sequentially numbered empty files in one command."
    },
    {
      "type": "technique",
      "signal": "actionable",
      "status": "definitive",
      "text": "Use tail -f to continuously monitor and display new lines appended to a log file in real time."
    },
    {
      "type": "technique",
      "signal": "actionable",
      "status": "definitive",
      "text": "Leverage history to quickly recall and re-execute the last five commands without retyping them."
    },
    {
      "type": "tool",
      "signal": "actionable",
      "status": "definitive",
      "text": "Use brace expansion in mkdir, cd, and touch commands to simplify path manipulation and file creation."
    }
  ],
  "tools_mentioned": [
    "mkdir",
    "cd",
    "touch",
    "tail",
    "history"
  ],
  "why_it_matters": "These Linux shortcuts streamline directory navigation and file management, directly boosting developer productivity.",
  "quality": 5,
  "transcript_source": "local-whisper",
  "transcript_chars": 976
}

curl

curl -X POST "https://api.apify.com/v2/acts/physealabs~reel-knowledge-distiller/runs?token=$APIFY_TOKEN" \
  -H "Content-Type: application/json" \
  -d @input.json

Python

from apify_client import ApifyClient

client = ApifyClient("YOUR_APIFY_TOKEN")
run = client.actor("physealabs/reel-knowledge-distiller").call(run_input={'reelUrls': ['https://www.youtube.com/shorts/fwBIZRq-vzY'], 'resultsLimit': 30, 'perspective': 'a research team tracking practitioner knowledge', 'llm': {'provider': 'openai-compatible', 'model': 'nvidia.nemotron-nano-3-30b', 'baseUrl': 'https://bedrock-runtime.us-east-1.amazonaws.com/openai/v1', 'apiKey': ''}, 'transcription': {'mode': 'local'}, 'minTranscriptChars': 80})
items = client.dataset(run["defaultDatasetId"]).list_items().items

Node.js

import { ApifyClient } from 'apify-client';

const client = new ApifyClient({ token: process.env.APIFY_TOKEN });
const run = await client.actor('physealabs/reel-knowledge-distiller').call({"reelUrls": ["https://www.youtube.com/shorts/fwBIZRq-vzY"], "resultsLimit": 30, "perspective": "a research team tracking practitioner knowledge", "llm": {"provider": "openai-compatible", "model": "nvidia.nemotron-nano-3-30b", "baseUrl": "https://bedrock-runtime.us-east-1.amazonaws.com/openai/v1", "apiKey": ""}, "transcription": {"mode": "local"}, "minTranscriptChars": 80});
const { items } = await client.dataset(run.defaultDatasetId).listItems();

You can call this Actor from an agent or LLM tool that can send HTTP requests to the Apify API. Keep the Apify token in a secret store.

FAQ

Does it open private posts?

No. It works with public links and does not log in to social platforms.

Can it process profiles?

It can list recent videos from public YouTube and TikTok profiles. Instagram profiles are skipped.

Can I use my own model?

Yes. Set the llm object with a compatible endpoint, model, and API key.