Skip to content

Inovagent/knowledge-gpt

BranchesTags

Repository files navigation

Knowledge GPT

Save information from supported sites into Notion or local Markdown without exposing secrets or filesystem paths in the browser. The project currently supports YouTube transcript capture and Gmail email-content capture, with room to expand further over time.

This repository currently contains:

  • a Chromium Manifest V3 extension that adds save buttons on supported pages like YouTube and Gmail
  • a small local Node.js API that writes captured content into Notion or backend-configured Markdown files

Why this project exists

Most browser-side Notion integrations force you to put a secret in the extension itself. This project avoids that by sending captured page data to a local backend, which then talks to Notion using your private integration token.

Features

  • Uses a local-first architecture so browser-side code never needs your Notion secret
  • Injects a save button into YouTube's watch-page action bar
  • Injects a Gmail-style save button into open Gmail message action bars
  • Adds right-click actions for clean web-page capture and selected-text capture
  • Opens the transcript panel automatically when needed
  • Extracts transcript text across multiple known YouTube DOM variants
  • Extracts readable email content from the active Gmail message body
  • Extracts readable page content with a Reader-style parser and a high-density page-section fallback
  • Opens an editable preview before saving generic page and selected-text captures
  • Sends captured data to a local backend for persistence into Notion or local Markdown
  • Creates a new Notion page or recreates an existing one based on your dedupe mapping
  • Writes local Markdown files with YAML frontmatter when localMarkdown is selected
  • Preserves a simple load-unpacked development flow

Repository structure

.
├── extension/
│   ├── manifest.json
│   ├── pages/
│   │   └── options/
│   ├── scripts/
│   │   ├── background.js
│   │   ├── content/
│   │   └── shared/
│   └── styles/
├── src/
│   └── backend/
│       ├── config/
│       ├── lib/
│       ├── middleware/
│       ├── routes/
│       └── services/
├── .env.example
└── package.json

Requirements

  • Node.js 18+
  • Chrome, Chromium, or another Chromium-based browser
  • A Notion integration with access to your target database, if saving to Notion
  • A local folder configured in the backend, if saving to local Markdown

Setup

1. Install dependencies

npm install
cp .env.example .env
bash scripts/install_hooks.sh

The hook installer adds the local Docs Impact git hooks for this clone: an advisory pre-commit reminder and a blocking commit-msg gate when documented hot paths change without a docs update or Docs-Impact: none - <reason> trailer.

2. Choose a storage destination

The extension requires an explicit destination before saving:

  • notion: writes captures into Notion using the backend's Notion token
  • localMarkdown: writes .md files into the backend folder configured by LOCAL_MARKDOWN_ROOT

Existing users must open the extension options page and choose a destination after upgrading.

3. Create a Notion integration

  1. Go to Notion integrations
  2. Create an internal integration
  3. Copy the integration token
  4. Share your destination database with that integration

Skip this step if you only plan to save local Markdown.

4. Configure the backend

Fill in .env:

  • NOTION_TOKEN: required for Notion saves, your Notion integration secret
  • HOST: optional, defaults to 127.0.0.1
  • PORT: optional, defaults to 8787
  • DEFAULT_NOTION_DATABASE_ID: optional fallback database ID for Notion saves
  • LOCAL_MARKDOWN_ROOT: required for local Markdown saves; must be an absolute path or start with ~/, point to an existing directory, and be writable by the backend

Start the server:

npm run start

The API will be available at http://127.0.0.1:8787 unless you change HOST.

Running the server in the background on macOS

If you want the backend to keep running without an open terminal, use a launchd agent.

  1. Create ~/Library/LaunchAgents/com.example.knowledge-gpt.plist
  2. Paste in:
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
  <dict>
    <key>Label</key>
    <string>com.example.knowledge-gpt</string>

    <key>ProgramArguments</key>
    <array>
      <string>/opt/homebrew/bin/node</string>
      <string>/path/to/knowledge-gpt/src/backend/server.js</string>
    </array>

    <key>WorkingDirectory</key>
    <string>/path/to/knowledge-gpt</string>

    <key>RunAtLoad</key>
    <true/>

    <key>KeepAlive</key>
    <true/>

    <key>StandardOutPath</key>
    <string>/path/to/knowledge-gpt/server.log</string>

    <key>StandardErrorPath</key>
    <string>/path/to/knowledge-gpt/server-error.log</string>
  </dict>
</plist>
  1. Load the service:
launchctl load ~/Library/LaunchAgents/com.example.knowledge-gpt.plist
launchctl start com.example.knowledge-gpt
  1. Verify that it is running:
launchctl list | grep knowledge-gpt
curl http://127.0.0.1:8787/health
tail -f "/path/to/knowledge-gpt/server.log"

Notes:

  • launchd restarts the backend if it exits
  • the backend will not run while the Mac is asleep
  • replace /path/to/knowledge-gpt with the actual path to your cloned repository
  • replace com.example.knowledge-gpt with your own preferred LaunchAgent label if you want
  • if node is installed somewhere else on your machine, run which node and update the plist path

Restarting the background server after code changes

If you change the backend source code, restart the launchd service to apply the changes:

launchctl unload ~/Library/LaunchAgents/com.example.knowledge-gpt.plist
launchctl load ~/Library/LaunchAgents/com.example.knowledge-gpt.plist
launchctl start com.example.knowledge-gpt

You can then confirm the new version is running with:

curl http://127.0.0.1:8787/health
tail -n 50 "/path/to/knowledge-gpt/server.log"

For local Markdown, the browser never sends a folder, filename, extension, or subpath. The backend generates direct-child .md filenames and writes with exclusive create semantics so existing files are not overwritten.

5. Load the extension

  1. Open chrome://extensions
  2. Enable Developer mode
  3. Click Load unpacked
  4. Select the repository's extension/ directory

The manifest pins a stable unpacked-extension ID with a public key, so saved options survive normal extension reloads and loading the extension from another local workspace path.

6. Configure the extension

Open the extension options page and set:

  • Storage destination: Notion or Local Markdown
  • Backend URL, for example http://127.0.0.1:8787
  • For Notion: Notion database ID, or leave it blank if you set DEFAULT_NOTION_DATABASE_ID
  • For Notion: property names for the fields you want this integration to populate
  • For Local Markdown: make sure /health reports that LOCAL_MARKDOWN_ROOT is set

After the extension is loaded, clicking the Knowledge GPT toolbar icon opens this options page directly.

Current scope

The current implementation supports:

  • YouTube watch pages for transcript capture
  • Gmail message views for email-content capture
  • Generic web pages through the right-click Save clean page action
  • Highlighted page text through the right-click Save selected text action

Generic page capture injects the capture script only after a user invokes the context menu. It uses Mozilla Readability against a cloned document to extract the main article-like content without page chrome, CSS, scripts, nav, and ads. If Readability cannot produce enough content, it falls back to scoring likely content containers such as article, main, and high-density page sections. Both generic flows show a preview dialog where the title, source, and content can be edited before saving.

Recommended Notion mapping

At minimum, configure:

  • a title property

Recommended optional mappings:

  • a URL property
  • an external ID property for stronger deduplication
  • a select property for Creator / Source
  • a select property for Source Type
  • a date property for the last synced timestamp The current YouTube flow writes the full transcript into the Notion page body as code blocks under a Transcript heading. The Gmail flow writes the extracted email body into the page body as code blocks under a Content heading. Generic page captures use an Article heading, and selected-text captures use a Selection heading.

Local Markdown output

Local Markdown files are named with this pattern:

YYYY-MM-DD-<contentType-or-sourceType>-<slug-title>-<short-externalId>.md

If a generated filename already exists, the backend writes -2, -3, and so on. Each file includes YAML frontmatter for available metadata and a Markdown body headed by Transcript, Content, Article, or Selection.

API

GET /health

Returns a simple status payload for the local backend, including booleans such as hasNotionToken, hasDefaultDatabaseId, and hasLocalMarkdownRoot. The local Markdown flag is only true when LOCAL_MARKDOWN_ROOT exists and is writable. It does not expose the configured local Markdown path.

POST /save-transcript

Accepts YouTube transcript saves using the shared capture schema.

Notion request:

{
  "storageDestination": "notion",
  "videoId": "abc123",
  "url": "https://www.youtube.com/watch?v=abc123",
  "title": "Example video",
  "channel": "Example creator",
  "transcript": "First line\nSecond line",
  "capturedAt": "2026-05-05T10:00:00.000Z",
  "databaseId": "optional-database-id",
  "propertyMapping": {
    "title": "Title",
    "videoUrl": "URL",
    "videoId": "External ID",
    "channel": "Creator / Source",
    "sourceType": "Source Type",
    "lastSyncedAt": "Last Synced At"
  }
}

POST /save-content

Accepts:

{
  "storageDestination": "localMarkdown",
  "externalId": "abc123",
  "url": "https://example.com/item/abc123",
  "title": "Example title",
  "source": "Example source",
  "sourceType": "Newsletter",
  "content": "First line\nSecond line",
  "contentType": "email",
  "capturedAt": "2026-05-05T10:00:00.000Z"
}

For Notion saves, include the Notion fields:

{
  "storageDestination": "notion",
  "externalId": "abc123",
  "url": "https://example.com/item/abc123",
  "title": "Example title",
  "source": "Example source",
  "sourceType": "Newsletter",
  "content": "First line\nSecond line",
  "contentType": "email",
  "capturedAt": "2026-05-05T10:00:00.000Z",
  "databaseId": "optional-database-id",
  "propertyMapping": {
    "title": "Title",
    "videoUrl": "URL",
    "videoId": "External ID",
    "channel": "Creator / Source",
    "sourceType": "Source Type",
    "lastSyncedAt": "Last Synced At"
  }
}

Development

Run the backend in watch mode:

npm run dev

Use npm run dev during active local development if you want the server to restart automatically when backend files change. Use the launchd setup above when you want the server to keep running in the background between terminal sessions.

Run syntax checks:

npm run check

Run tests:

npm test

Run the documentation guardrails directly:

bash scripts/docs_impact_guard.sh --worktree
bash scripts/docs_consistency_check.sh

Specs

Durable product and implementation specs live in specs/. Start with specs/README.md for the format and specs/0001-capture-and-storage-platform.md for the current capture and storage platform contract.

Publishing notes

Before publishing publicly, you will probably still want to add:

  • a screenshot or demo GIF
  • a real repository URL in package.json
  • GitHub issue templates or contribution automation if you want outside contributors

License

MIT

About

Simple Extension to save youtube transcripts in notion

Resources

Readme
MIT license

Contributing

Contributing
Activity
Custom properties

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages