YouTube Watch

What it is

A content-intelligence pipeline built on YouTube. You (or an agent) mark a set of channels to watch; the agents then poll those channels for new videos, pull each video’s transcript, count and language-tag it, run an AI enrichment pass over it, and finally digest the result into your Knowledge base. It is how market signal from video gets turned into something your agents can read. The pipeline has two tables, documented below: Watched Channels (youtube_channels) — the sources you follow — and Videos (youtube_videos) — the pulled, transcribed, and enriched content.

How it gets data

Populated by the market-intelligence agents that watch channels and process videos; not part of the standard onboarding catalog. Available to query once your workspace runs the relevant agents.

Fields — Watched Channels (`youtube_channels`)

Field	Type	Description
`external_id`	text	Source channel id — the dedup/upsert key
`channel_name`	text	Channel name
`channel_url`	text	Link to the channel
`topic`	text	What this channel is watched for
`watch_enabled`	boolean	Whether the pipeline is actively polling this channel
`last_checked_at`	datetime	When the channel was last polled for new videos
`last_video_id`	text	The most recent video seen on this channel
`last_published_at`	datetime	Publish time of that most recent video
`notes`	text	Free-text notes on the channel
`metadata`	json	Raw source fields kept alongside the row
`digest_knowledge_id`	uuid	The Knowledge entry this channel’s digest feeds

Fields — Videos (`youtube_videos`)

Field	Type	Description
`external_id`	text	Source video id — the dedup/upsert key
`channel_external_id`	text	The watched channel this video came from
`channel_name`	text	Channel name (denormalized for convenience)
`channel_topic`	text	Channel topic (denormalized for convenience)
`title`	text	Video title
`video_url`	text	Link to the video
`published_at`	datetime	When the video was published
`duration`	text	Video duration
`transcript_full`	text	Full transcript text
`transcript_language`	text	Detected transcript language
`word_count`	integer	Transcript word count
`ai_enrichment`	json	The AI-enrichment output (summary, entities, signals)
`enrichment_status`	text	Pipeline stage (e.g. pending, enriched, failed)
`enriched_at`	datetime	When enrichment finished
`enrichment_model`	text	Model used for enrichment
`digested_at`	datetime	When the video was digested into Knowledge
`metadata`	json	Raw source fields kept alongside the row

Both tables also carry the system columns id, tenant_id, workspace_id (tenant isolation), created_at, and updated_at. They are managed automatically.

Query it

{
  "method": "tools/call",
  "params": {
    "name": "data_query",
    "arguments": {
      "view_slug": "youtube_videos",
      "filters": { "enrichment_status": "enriched" },
      "select_fields": ["title", "channel_name", "published_at", "word_count", "video_url"],
      "sort_by": "published_at",
      "limit": 50
    }
  }
}

AI Operating Guide

Vertical Intelligence

Getting Started

Governance

Resources

Sales & Commerce

Procurement

Marketing & Outreach

What it is

How it gets data

Fields — Watched Channels (`youtube_channels`)

Fields — Videos (`youtube_videos`)

Query it

​What it is

​How it gets data

​Fields — Watched Channels (youtube_channels)

​Fields — Videos (youtube_videos)

​Query it

What it is

How it gets data

Fields — Watched Channels (`youtube_channels`)

Fields — Videos (`youtube_videos`)

Query it