# n8n YouTube agent: keep the transcript behind the answer Use the long tutorial as a staged route from agent setup to a reusable, source-backed video workflow. [HTML](https://lukejbyrne.com/library/guides/k4iargzlhta/) Content language: en Author: Luke Byrne Video published: 2025-02-25 Guide published: 2026-11-02 [Watch the video](https://www.youtube.com/watch?v=K4iaRgZlhtA) Video language: English ## Build the agent core before adding a collection The 25 February 2025 tutorial describes an n8n agent for video questions, summaries, newsletters, descriptions and key timestamps. It combines OpenAI, memory, tools, a community transcription node and a database. Those outputs all depend on whether the intended video material actually reached the agent, so begin with one identifiable source rather than a large library. 1. Use the retained n8n route: https://n8n.partnerlinks.io/trxvmi48ria4. 2. Follow the core at 05:35 and identify the model, memory and tool roles in the demonstrated workflow. 3. Choose one video and a question whose answer you can check from its content before adding further sources. ## Inspect transcription and database preparation The community transcription node begins at 14:53 and formatting/database work at 22:16. A retrieved transcript and a successfully stored record are distinct checkpoints. Preserve the source identity when formatting text so a later answer or timestamp does not become detached from the video it describes. 1. Follow the recorded community-node setup; the description does not provide an exact package name or installation command. 2. Inspect the actual transcript for missing passages or unavailable content before sending it downstream. 3. Check formatted text and the database record against the source video, keeping identifiers and time information where the workflow uses them. 4. Distinguish an empty retrieval or failed write from a completed ingest rather than treating every run as a usable new source. ## Connect retrieval and verify the answers The database-reading stage begins at 34:00. Memory and retrieval should return the intended source, not merely produce a plausible recollection. The description says the agent can fact-check videos, but agreement with a transcript only establishes what the video says; verifying its claims may need independent evidence. 1. Ask the initial known-answer question and compare the response with the transcript passage it depends on. 2. Request a key moment and verify the proposed timestamp against the recording rather than accepting a generated number. 3. Keep unsupported statements, ambiguous sources and missing content visible when reviewing a summary or newsletter draft. ## Open the generated file and preserve its source trail File generation begins at 39:14 and the recap at 47:06. A file that exists can still omit qualifications, mix videos or lose the requested structure. Review its content before treating it as a publishing handoff or finished resource. 1. Open the delivered file and compare its title, claims and sections with the intended video and request. 2. Save the source association and unresolved checks with the output so later edits can be traced. 3. Check current trial and provider costs separately; the recorded 14-day trial is historical, not a promise of a free ongoing system. ## The takeaway The useful video agent retrieves recognisable source material and produces answers or files that can be checked against it. ## Workflow resources - [n8n.partnerlinks.io](https://n8n.partnerlinks.io/trxvmi48ria4) - [ailuke.short.gy](https://ailuke.short.gy/K4iaRgZlhtA1) - [ailuke.short.gy](https://ailuke.short.gy/K4iaRgZlhtA2) - [datacamp.pxf.io](https://datacamp.pxf.io/yqr3jb) - [newsletter.lukejbyrne.com](https://newsletter.lukejbyrne.com) - [linkedin.com](https://linkedin.com/in/lukejbyrne) - [datacamp.pxf.io](https://datacamp.pxf.io/gO6WRB) - [datacamp.pxf.io](https://datacamp.pxf.io/XmnX4a) - [datacamp.pxf.io](https://datacamp.pxf.io/Bny25L) - [datacamp.pxf.io](https://datacamp.pxf.io/aOAWNZ) - [datacamp.pxf.io](https://datacamp.pxf.io/AP3Eg1) - [datacamp.pxf.io](https://datacamp.pxf.io/GKVZbk) - [datacamp.pxf.io](https://datacamp.pxf.io/Z6KyV1) ## Chapters - [00:00 - Introduction to AI-Powered YouTube Chat & n8n](https://www.youtube.com/watch?v=K4iaRgZlhtA&t=0s) - [05:35 - Building the AI Agent Core: OpenAI, Memory, & Tools](https://www.youtube.com/watch?v=K4iaRgZlhtA&t=335s) - [14:53 - Unleashing Community Power: The YouTube Transcription Node](https://www.youtube.com/watch?v=K4iaRgZlhtA&t=893s) - [22:16 - Preparing Transcript Data: Formatting and Database Setup](https://www.youtube.com/watch?v=K4iaRgZlhtA&t=1336s) - [34:00 - Connecting the Agent to the Database and Reading from it](https://www.youtube.com/watch?v=K4iaRgZlhtA&t=2040s) - [39:14 - Generating Files Directly from the AI Agent](https://www.youtube.com/watch?v=K4iaRgZlhtA&t=2354s) - [47:06 - Conclusion: Recap & Community Invitation](https://www.youtube.com/watch?v=K4iaRgZlhtA&t=2826s) ## Sources and limitations Prepared 2 November 2026 from this exact published description and chapters. The recording was not independently watched or transcribed and no fresh hands-on test was performed. Steps are editorial adaptations; current setup, availability and offer terms may differ. The source spells its database service “Superbase”; the exact service configuration and schema are not reproduced in the published description. [Original video](https://www.youtube.com/watch?v=K4iaRgZlhtA)