Skip to main content

Documentation Index

Fetch the complete documentation index at: https://koreai.mintlify.app/llms.txt

Use this file to discover all available pages before exploring further.

Back to Search AI connectors list YouTube is a video-sharing platform that allows users to upload, view, and manage video content organized into channels and playlists. You can configure Search AI to connect to YouTube to enable users to fetch query results using video metadata, descriptions, and transcripts from your YouTube channel.
SpecificationDetails
Repository typeCloud
Supported contentVideos (metadata, descriptions, transcripts)
Content filteringYes (playlist selection and advanced filters)
RACL supportYes (available; can be enabled)

​
Authorization Support

Search AI supports OAuth 2.0 (Google) authentication for communicating with YouTube. Authentication requires a Client ID and Client Secret obtained from the Google Cloud Console.

​
Integration Steps

To configure YouTube as a content source, complete the following steps.
  • Configure the YouTube connector in Search AI
  • Set up Permissions
  • Configure Content Scope
  • Schedule Sync

​
Step 1: Configure the YouTube Connector in Search AI

​
Prerequisites - Obtaining Google OAuth 2.0 Credentials

Before configuring the connector, generate your OAuth 2.0 credentials from the Google Cloud Console:
  1. Go to Google Cloud Console and create or select a project.
  2. Enable the YouTube Data API v3 under APIs & Services > Library.
  3. Configure the OAuth consent screen (set user type to Internal for org use).
  4. Add the required scopes:
    • https://www.googleapis.com/auth/youtube.readonly - View your YouTube account
    • https://www.googleapis.com/auth/youtube.force-ssl - Required for downloading captions/transcripts
  5. Create an OAuth client ID (Web application type) under APIs & Services > Credentials.
  6. Add the authorized redirect URI: https://<your-searchassist-domain>/searchassist/idproxy/callback
  7. Copy the generated Client ID and Client Secret.

​
Configuring the Connector

Go to Connectors under the Sources page and select YouTube. On the Authentication tab, enter the following details and click Connect.
FieldDescription
NameA unique name for this connector instance.
Authorization typeOAuth 2.0 (pre-selected).
Grant typeAuthorization Code (pre-selected).
Client IDYour Google OAuth 2.0 Client ID (for example: 123456789-abcdef.apps.googleusercontent.com).
Client SecretYour Google OAuth 2.0 Client Secret (for example: GOCSPX-xxxxxxxxxxxxxxxxxxxxxx).
After entering credentials and clicking Connect, you will be redirected to Google’s consent screen to authorize Search AI. Tokens are stored securely and refreshed automatically when they expire. Once successfully authenticated, the status shows as Connected.

​
Step 2: Set Up Permissions

Go to the Permissions tab to configure access control for the ingested content. The following options are available.
  • Same users as in the source system (Restricted Access) - Applies RACL-based access control. When enabled, the connector uses the permissions field from the video data source. Since YouTube videos are inherently public (private videos are skipped), RACL effectively defaults to open access for all ingested content.
  • Everyone (Public Access) - All users are granted full access to the ingested content.
RACL Sync scheduling isn’t applicable for the YouTube connector since YouTube doesn’t have a user/group permission model for public video content.

​
Step 3: Configure Content Scope

Go to the Content Scope tab under Connector Setup to define which videos are ingested. Content scope is controlled via two levels of filtering.

​
Standard Filter (Playlist Selection)

The standard filter controls which playlists to sync content from.
OptionBehavior
Sync All (default)Syncs videos from all playlists in your channel. If no playlists exist, falls back to the channel’s “Uploads” playlist (all uploaded videos).
Select Specific PlaylistsDisplays a playlist picker where you can search and select specific playlists. Only videos from the selected playlists are synced.
Select All PlaylistsDiscovers and syncs all playlists from the channel (same as Sync All, but explicit).
The Playlist Picker supports paginated browsing, search/filter by name, and multi-select.

​
Advanced Filter (Pre-Filter)

Advanced filters let you narrow down which videos are ingested based on specific criteria. These filters are applied before transcript fetching, so filtered-out videos don’t consume API quota.
Filter ParameterDescriptionExample Value
publishedAfterOnly include videos published after this date.2024-01-01 or 2024-01-01T00:00:00Z
publishedBeforeOnly include videos published before this date.2025-12-31
minViewCountOnly include videos with at least this many views.1000
minDurationMinutesOnly include videos longer than this (in minutes).5
maxDurationMinutesOnly include videos shorter than this (in minutes).60
Validation Rules:
  • publishedAfter must be earlier than publishedBefore
  • minDurationMinutes must be less than maxDurationMinutes
  • View count and duration values must be non-negative
  • Dates accept ISO 8601 format (YYYY-MM-DD or YYYY-MM-DDTHH:MM:SSZ)
Example: “Only sync tutorial videos published after January 2024 that are between 5 and 60 minutes long and have at least 500 views.”

​
Step 4: Sync Content

After configuring the connector, go to Schedule Sync to initiate and manage content synchronization.
  • Use the Sync button for an immediate, on-demand sync.
  • Enable Schedule Sync to set up automated, recurring syncs.
The Schedule Sync table displays a log of all sync activity with the following details.
FieldDescription
Sync ScopeThe scope of the sync (for example, Full Sync).
Sync TypeIndicates whether the sync was On-Demand or Scheduled.
Sync StatusThe result of the sync (for example, Successful).
Triggered ByThe user who initiated the sync.
Started OnThe date and time the sync started.
Completed OnThe date and time the sync completed.

​
Sync Behavior

AspectBehavior
Full SyncFetches all videos from the configured playlists, processes them, and updates the index. A delete queue runs after ingestion to remove any previously indexed videos that are no longer present.
SchedulingYou can configure an automatic sync schedule to keep content up to date.
Shorts HandlingYouTube Shorts (videos under 60 seconds) are excluded by default.
Only public videos are synced. Private and unlisted videos are automatically skipped.

​
Content Ingestion

Once the sync is complete, go to the Content tab to review the ingested content. The tab displays the count of files that are Successful, Failed, and Skipped, along with the total number of Accessible Files. Each synced video is converted into a single markdown document. The following key fields are extracted and indexed.
  • title - The video title.
  • content - The full transcript organized by chapters (if available) with timestamps, plus the first 2,000 characters of the video description.
  • doc_id - A unique identifier for the ingested video document, derived from the YouTube video ID.
  • doc_source_type - Identifies the source as a YouTube video.
  • doc_created_on / doc_updated_on - Timestamps capturing when the video was published and last updated.
  • url - Direct link to the video on YouTube (https://www.youtube.com/watch?v=<videoId>).
  • tags - Video tags extracted from YouTube metadata.
  • type - Content type classification (for example: tutorial (85%)).

​
Content Format

Each ingested video document follows this structured layout: 
# <Video Title>

**Channel:** <Channel Name> | **Published:** <Date> | **Duration:** <HH:MM:SS>
**Video:** https://www.youtube.com/watch?v=<videoId>
**Tags:** tag1, tag2, tag3
**Type:** tutorial (85%)

## Description

<First 2,000 characters of the video description>

## Chapter Title 1 (0:00)

[00:00] First transcript line...
[00:05] Second transcript line...
[00:12] Third transcript line...

## Chapter Title 2 (3:45)

[03:45] Transcript continues here...
[03:50] More transcript text...

​
Transcript (Captions) Behavior

The connector fetches video captions/transcripts using the YouTube Captions API. Transcript selection follows this priority order:
PriorityTranscript Type
1stManual (human-uploaded) captions in English
2ndAuto-generated captions in English
3rdManual captions in any language
4thAuto-generated captions in any language
  • If a video has no captions at all, it’s still synced - the video metadata and description are indexed without a transcript.
  • Chapter markers in the video description (for example: 0:00 Introduction, 5:30 Setup) are automatically detected and used to organize the transcript into sections.

​
API Quota

The YouTube Data API v3 has a daily quota limit of 10,000 units per Google Cloud project. The connector is designed to be quota-efficient:
OperationQuota Cost
List channels / playlists / videos1 unit per request
List captions for a video50 units per request
Download a caption track (transcript)200 units per request
The connector reserves a buffer of 1,000 units and stops processing if the quota budget (9,000 units) is reached. If quota is exhausted mid-sync, the job completes with a “limit exceeded” status and already-processed videos are retained.
Tip: If you have a large channel (hundreds of videos with transcripts), consider requesting a quota increase from Google through the Cloud Console Quotas page.

​
RACL Support

Search AI provides access control support for content ingested from YouTube. The sys_racl field is used to enforce access control for the ingested content.
FeatureStatus
RACL SupportAvailable (can be enabled)
RACL SyncNot applicable
  • When RACL is disabled (default), all synced content is publicly accessible to all users searching the index (permission set to *).
  • When RACL is enabled, the connector uses the permissions field from the video data source. Since YouTube videos are inherently public (private videos are skipped), RACL effectively defaults to open access (['*']) for all ingested content.
  • RACL Sync scheduling isn’t applicable for the YouTube connector since YouTube doesn’t have a user/group permission model for public video content.