Verdalia
← Back to Guides

Media Tracker with Tables and Documents#

This guide uses a simple media tracker as the pattern: structured storage holds videos and links, while documents hold notes, summaries, and decisions.

Use this for YouTube videos, podcasts, talks, tutorials, papers, or anything else you want to review over time.

What the table declares#

A table is the structured half: named columns, each with a declared kind, and one row per thing you are tracking. What has to be decided is which columns one of those rows has, what kind each column is, and which parts are prose that belongs in a document instead.

Create a media-watchlist table for repeated media. Each row is one thing to watch, read, or review.

A starter declaration might use these columns:

Column Kind Use
title Text Video, episode, or article title
url Text Source link
channel Text Creator, publication, or source
status Select queued, watched, summarized, archived
priority Integer What to review first
topic Text Main subject
notes_doc Resource The related notes document
source JSON The raw feed payload the ingest workflow kept

status as a Select is what stops four spellings of "watched" accumulating over a year of writes. notes_doc is a Resource pointed at Document, so the row stores the document's id rather than a slug you would have to keep in step with renames.

Messy API output — the raw YouTube or RSS payload — is a JSON column. Everything else in the set is checked on write, so declaring source as JSON is how you say that this one is not, and a workflow can write and read it back without anyone pretending it has a shape.

topic starts as Text because the subjects are open. If yours settle into a fixed set, retyping it to a Select is instant — Text and Select are stored the same way, so nothing is rewritten. Tables and Agent Tools covers the rest of the kinds and what a change to a declaration costs.

What goes in documents#

Use documents for the human-readable part:

  • watch notes
  • summaries
  • timestamps
  • quotes
  • follow-up questions
  • topic digests
  • weekly review documents

A document can explain why a video mattered. The row keeps it findable.

A common workflow#

RSS or YouTube source → Table → AI summary → Notes document → Memory

The table is the queue and index. The document is the review artifact.

Example flow#

  1. Ingest new videos from YouTube RSS, generic RSS, Reddit, Hacker News, or a web search.
  2. Store each one as a row in the media-watchlist table.
  3. Use AI to classify the topic and estimate priority.
  4. When a row is watched, create a notes document.
  5. Save durable takeaways to memory so the assistant can recall them later.

A bulk ingest usually writes with a Table node, one batch per run rather than one row at a time. Those writes are checked against the declaration all the same: a missing title, a status nobody declared, a fraction in priority — each is refused rather than stored, so a malformed feed fails at ingest instead of a month later. Narrowing the status options while rows still hold the one you are removing is refused too, and names the row.

What can still read as broken is a row that predates the declaration it is now read against — an earlier shape of the table, or data brought in from elsewhere. It shows in place with the reason and a repair beside it, and counts in every total.

Useful assistant prompts#

Find queued videos about AI coding and sort them by priority.

Create a watch notes document for this video.

Summarize the watched videos from this week into a short digest.

Which topics keep appearing across my saved videos?

Why this pattern works#

Media queues get messy quickly. The table keeps the queue structured and typed: status, source, topic, priority. Documents keep the thinking readable.

Verdalia keeps both connected. A workflow can create rows, update statuses, generate notes, and feed durable takeaways into memory.

Good habits#

  • Keep one row per media item.
  • Let status be a Select rather than Text.
  • Store long notes in documents, not in Long text columns.
  • Point at a document with a Resource column rather than pasting its slug into Text.
  • Declare an API-shaped payload as JSON instead of flattening it too early.
  • Use documents for weekly digests or topic summaries.
  • Save only durable takeaways to memory.