Use cases

Turn any content into structured data

Name the fields you need and describe what belongs in each one. Owlie reads every new video, article, podcast episode or post from the sources you follow and returns JSON in exactly that shape, every time.

The recipe

One Operation does the whole job. These are the settings to choose when you create it; you can change any of them later.

  1. Sources

    Add what you want to read: a YouTube channel or a single video, an RSS feed or a single article, a podcast feed, or a subreddit. One Operation can follow several Sources.

  2. Filter

    Choose Everything to structure every new item, or structure only the items that matter by narrowing them first with keywords or with criteria written in plain language for the AI.

  3. Transform

    Choose Structured content and add each field: a name, a type, and a description of what to extract. The fields are the whole instruction, so describe each one clearly.

  4. Deliver

    Every Result lands in your Inbox. Add a webhook to receive each one as JSON the moment it is ready, or let your own tools and AI agents read Results over MCP.

Example

Illustrative example based on the Owlie CLI v0.6.0 release notes. Each field description says what to extract, and the Result fills every field in that shape.

  1. Extracted content

    Text Owlie extracted from a GitHub release page.

    github.com/mihaiiova/owlie/releases

    Owlie CLI v0.6.0
Released Oct 1, 2026

Added
- owlie extract <page URL> returns the article when the page is one.
- One proxy for all extraction traffic, set with OWLIE_PROXY_URL.
- YouTube /embed/<id> and /v/<id> URLs.

Changed
- Owlie now requires Node.js 22 or later (was 20).
- YouTube /live/ and /shorts/ URLs fail with VALIDATION_ERROR.

Fixed
- A fetch stopped by a timeout no longer crashes the process.

  2. Fields

    Each field has a name, a type and a description of what to extract. The fields are the whole instruction.

    Fields: what to extract

    {
  "version": {
    "type": "text",
    "description":
      "Released version number"
  },
  "summary": {
    "type": "text",
    "description":
      "What changed, in one sentence"
  },
  "breaking_changes": {
    "type": "list of text",
    "description":
      "Changes that can break setups"
  },
  "new_features": {
    "type": "list of text",
    "description":
      "New capabilities, briefly"
  },
  "upgrade_recommended": {
    "type": "boolean",
    "description":
      "Is it worth upgrading now?"
  }
}

  3. Structured Result

    Each field filled as described, ready for your Inbox, a webhook or MCP.

    Result

    {
  "version": "0.6.0",
  "summary":
    "Adds article pages and a proxy",
  "breaking_changes": [
    "Requires Node.js 22 or later",
    "No YouTube /live/ or /shorts/"
  ],
  "new_features": [
    "Article extraction from URLs",
    "One proxy for all traffic",
    "YouTube /embed/ and /v/ URLs"
  ],
  "upgrade_recommended": true
}

Variations

The same recipe with different Sources and fields.

  • Track releases of the tools you use

    Follow a project’s GitHub releases feed and get the version, breaking changes and an upgrade recommendation for every release, as in the example above.

  • Build a dataset from news articles

    Follow a publication’s RSS feed and extract the same columns from every article: companies, people, amounts, dates. A webhook can write each row straight into your database or spreadsheet.

  • Turn talks and interviews into notes

    Follow a YouTube channel or a podcast and capture the topic, the tools mentioned, the key claims and any recommendations from every new video or episode.

  • Classify community posts

    Follow a subreddit and label each post with an enum field (bug, question, feature request) and a boolean such as “mentions a competitor”, so you can count and route them.

Good to know

What this recipe can and can’t do today.

  • Fields are flat

    Each field is text, a number, true or false, a URL, a date, one of a fixed set of choices, or a list of simple values. Nested objects are not supported. An Operation can have up to 40 fields.

  • Missing values are null

    When an item doesn’t contain what a field asks for, that field is null instead of a guess. Clear field descriptions make this rare.

  • The AI reads the content, not the page address

    Owlie gives the AI the item’s content, its title and the Source name. Don’t add a field for the source URL: every Result already links to its original.

  • Some YouTube formats are not supported

    Live streams and Shorts can’t be read. Regular videos work, from a channel or on their own.

  • What it costs

    Each structured Result costs 1 credit plus its AI usage, so at least 2 credits. Podcast audio adds transcription at 0.25 credits per minute. Items a keyword filter leaves out cost nothing. New accounts start with free credits.

Questions

Short answers to what people ask most.

  • Can I define my own JSON schema?

    Yes. You name every field, choose its type and describe what to extract. Choice fields take a fixed list of values, and number fields can have a minimum and maximum. Owlie checks each Result against your field types.

  • Does it work with YouTube videos and podcasts, not only articles?

    Yes. Owlie extracts the text first: an article’s body, a video’s transcript, or a podcast episode’s audio, transcribed. The same fields then work for any of them.

  • How do I get the JSON into my own app?

    Add a webhook to the Operation and Owlie sends each new Result as JSON to your URL. AI agents and other tools can also read your Results over MCP with a personal access token.

  • Does it run on new content automatically?

    Yes. An Operation keeps watching its Sources and turns every new item into a Result. You can also run it once on a single video or article.

  • Is the code that reads web pages open source?

    Yes. Owlie’s extractor is the open-source Owlie CLI, which you can also run on your own machine.

More use cases

Other ways to put Owlie to work.

Set this up in a few minutes.

Create an account, add a source, and choose the fields Owlie extracts from every new item.

Create accountSign in