OXYGENOXYGEN

Public web into rows

Scrape the web into typed rows, not a CSV dump

Scraping is only a problem because of what happens after it. Here a scrape is a column on a table: one row per page, one cell per field, with retries and errors visible instead of buried in a log.

try
Start freeRead the docs

The scraper is rarely the hard part. Hosting it is, and so is the parser that breaks when a page changes, the retry logic nobody documented, the proxy that gets blocked on a Friday, and the question of where the output goes once it finally exists.

A scrape here is a tool column. The input is a URL or a domain already sitting on the row, the output is a typed cell, and the run state belongs to the table: which pages failed, what each one drew, and what the extractor actually saw. Rerunning just the failures is one action rather than a script.

Example searches

Ask for it the way you'd say it

Every search below runs on the same hosted Tables, with the cost previewed before anything is spent.

  • Read a page per account

    open each company's pricing page and tell me whether they publish a price at all

    One cell per row holds the answer and the page it came from, so a wrong extraction stays checkable.

  • Turn a directory into rows

    crawl this marketplace listing and give me every vendor with a website and a category

    A crawl writes rows rather than a file, so enrichment can start on the same pass that found them.

  • Watch a page for changes

    check these competitor changelogs every week and flag anything that mentions an integration

    A scheduled rerun leaves last week's value on the cell with the run that wrote it, so the difference is readable.

Plays

Three motions you can run this week

Each one is a chain of Oxygen primitives — the same hosted objects your workspace already has, composed.

  • Pricing pages into a comparison column

    Every account in the table carries a structured read of what that company publicly charges.

    1. Tables
    2. Knowledge Graph
    1. 1Add a scrape column keyed on the domain, then an extraction column that pulls the fields you need.
    2. 2File the pattern you found as workspace knowledge so later runs read it instead of rediscovering it.
    oxygen columns add

  • A public directory becomes a table

    A listing of companies turns into rows you can enrich, score, and eventually promote.

    1. Tables
    2. Records
    1. 1Crawl the listing pages into rows, then collapse on domain before spending anything on them.
    2. 2Promote the ones that clear your filters and leave the rest as a table you can revisit.
    oxygen tables dedupe

  • Rescrape on a schedule

    A page you care about is reread on a cadence and the difference is visible on the row.

    1. Tables
    2. Workflows
    3. Messages
    1. 1Schedule the column with a per-delivery ceiling so a large table cannot surprise the balance.
    2. 2Compare the new cell against the run before it, then let a workflow read the table and decide whether the change deserves a message.
    oxygen tables schedule set

Capabilities

What you get

  • No crawler to operate

    Fetching, retries, and blocking are the provider's problem. What you configure is the input column, the extraction shape, and the ceiling on the run.

  • Failures you can see and retry

    A page that did not load leaves an error on its own cell. Running the column again over just those rows is one scoped command, and the rows that already answered are left alone.

  • Extraction as a typed column

    The output has the shape you declared, so downstream columns and filters rely on it instead of parsing free text for a second time.

  • Scrapes that feed the rest

    Because the result is a cell on a row, scoring, enrichment, and enrollment read it directly. There is no handoff file between the scrape and the campaign.

Boundaries

What Tables does not own

Scraping is a Tables column, not an orchestration engine. Scheduling, branching, and retries across many steps belong to Workflows. A page change you want to act on later is a value you compare between runs, and the run history that makes that possible is Tables.

Data sources

What the data actually comes from

Every value lands with its provider and cost recorded on the cell.

  • Firecrawl

    The workhorse for a page or a whole site: one URL in, structured content out, plus the crawl and batch paths when a traversal is genuinely needed.

  • Linkup

    A cheap managed fetch to clean markdown, and the second lane when the first one comes back with nothing usable.

  • Exa

    A different fetch substrate, a cached index with a live crawl behind it, which is why it answers on pages where the live fetchers are blocked.

Run these on Oxygen's managed credits, or connect your own provider keys and pay the vendor directly — the same columns, the same runs, the same provenance either way. See every integration.

Limits

Where this stops

  • Public pages change without warning and some sites decline to be read at all. Treat an empty cell as a fact about the page, and check a sample by hand before trusting a field across thousands of rows.

FAQ

Questions people ask first

Do I need to host anything?
No. Fetching, retries, and blocking are handled upstream. You configure the input column, the extraction shape, and the ceiling, and the run state stays visible inside the table.
What happens when a page fails to load?
The cell records the error and the row stays where it is. Rerunning is scoped to the rows you name, so a partial run is never a reason to start the whole table again, and the cells that already answered keep the cost you paid for them.
Can I crawl an entire site?
You can crawl a set of pages into rows. It is usually cheaper to find the exact page first and read that, because most of any site has nothing to do with the question you asked.
Is the output free text or structured?
You declare the shape you want and the column writes into it. Downstream filters and scoring then read typed cells rather than reparsing the same paragraph over again.
How is a scrape priced?
Per call, drawn in credits, and previewed on your own rows before the run starts. Keeping the column off rows that already hold an answer is the cheapest way to bring that number down.
Is scraping allowed?
That depends on the site and on your jurisdiction, and the call is yours to make. Oxygen reads public pages through providers; it does not advise you on whether a particular source should be read.

Run scrape the web into typed rows, not a csv dump on your own workspace

Sign up, get a working workspace with a one-time credit grant, and run the first play from the web app, the CLI, or your AI assistant over MCP.

Start freeSee pricing