Content Auditor Agent

Watches/reads/listens to video, image, and audio files, extracts the main points as bullets, flags inconsistencies/gaps/referenced-but-unexplained topics, pushes a note to Notion per file, and runs a weekly audit that reports which notes you never actually touched.

It never edits or rewrites your source material — it only reports.

1. Get a Gemini API key

Go to https://aistudio.google.com/apikey and create a key. Free tier is fine to start.

2. Create a Notion integration

1. Go to https://www.notion.so/my-integrations → New integration
2. Name it (e.g. "Content Auditor"), pick your workspace, save
3. Copy the Internal Integration Token (starts with secret_ or ntn_)

3. Create the Notion database

Create a new database (as a full page in Notion) with exactly these properties:

Property name	Type
Name	Title (default)
Source	Text
Type	Select
Date Added	Date
Used	Checkbox

You don't need to pre-fill the Select options — the agent creates Video / Image / Audio / Audit options automatically the first time each one is used.

Then click ... on the database → Connections → connect the integration you just created. (This step is easy to miss — without it the API gets a 404 on every request.)

4. Get the database ID

Open the database as a full page and copy the ID out of the URL:

https://www.notion.so/myworkspace/8a1b2c3d4e5f6789...?v=...
                                   ^^^^^^^^^^^^^^^^^^ this part (32 chars)

5. Install and configure

cd ai-notes-agent
pip install -r requirements.txt
cp .env.example .env

Edit .env and fill in:

GEMINI_API_KEY=...
NOTION_TOKEN=...
NOTION_DATABASE_ID=...

6. Run it

Drop video/image/audio files into the inbox/ folder, then:

python ingest.py

Each file gets analyzed, a note gets created in Notion, and the file moves to processed/ so it's never re-analyzed. A processed_log.json tracks what's been done.

Supported formats:

• Video: mp4, mov, avi, webm, mkv
• Image: jpg, jpeg, png, webp, gif
• Audio: mp3, wav, m4a, aac, ogg, flac, aiff

7. Or run it in watch mode (live agent)

Instead of running ingest.py by hand every time, leave this running in a terminal:

python watch.py

It polls the inbox/ folder every 3 seconds and auto-processes any new video/image/audio file the moment it's fully copied in — waits for the file size to stop changing first, so it won't grab a video mid-copy. Ctrl+C to stop. Good for demos: drop a file in, watch the note appear in Notion without touching the keyboard again.

8. Run the weekly audit

python audit.py

This looks at every note added in the last 7 days and reports:

• Which notes show no sign you opened/edited them since creation (unused)
• Topics referenced across 2+ sources this week that are still unexplained
• Every inconsistency and gap flagged that week

It prints the report as bullets to the terminal and creates a "Weekly Audit" page in the same Notion database.

Automating the weekly run

Mac/Linux (cron) — run every Sunday at 6pm:

crontab -e
# add this line:
0 18 * * 0 cd /full/path/to/ai-notes-agent && /usr/bin/python3 audit.py >> audit.log 2>&1

Windows (Task Scheduler): create a weekly trigger that runs python audit.py with "Start in" set to the project folder.

How "used" is detected

Notion pages track created_time and last_edited_time automatically. If a note's last_edited_time still equals its created_time, nothing has touched it since the agent created it — it's marked unused. You can also manually tick the Used checkbox on any note to mark it used regardless. Note: this only catches edits, not views — opening a page without changing anything won't register as "used."

Notes on cost/limits

• Gemini's free tier has per-minute/per-day request limits — fine for a contest demo, may need a paid key for heavy volume.
• Large video/audio files go through Gemini's Files API automatically and can take a few seconds to a couple minutes to finish processing before analysis starts — this is normal, the script waits for it.