The perception layer for LLMs: webpages in, structured records out. Deterministic, zero-shot, honest.

The internet was made for human eyes. Analog turns rendered webpages into structured records that AIs can work with directly: queryable, exportable, and diffable. Extraction is deterministic and uses no LLMs.

The SDK fetches webpages in a local browser by default, then sends the URL and rendered page content to Analog for structured extraction. Results return to the SDK and are saved locally unless disabled.

The default browser visits as an unauthenticated user. Analog respects robots.txt and stays recognizable and blockable: no stealth, proxies, fingerprint spoofing, or CAPTCHA solving. mode="local" is the Markdown-only path that sends nothing.

analog get https://books.toscrape.com/

Selected fields from one extracted record:

{
  "product": "A Light in the Attic",
  "price": "£51.77",
  "rating": "3/5",
  "badge": "In stock"
}

Analog is currently in private alpha. Join the private alpha →

Already have access? Install Analog · Run the quickstart · Use Analog through MCP

• analog-sdk — the analog command, Python SDK, and built-in browser.
• analog-mcp — Analog for MCP-compatible agents.
• homebrew-tap — Homebrew distribution for the Analog CLI.

• marcato — deterministic HTML-to-Markdown conversion on lxml.

• Agentic delight. We love building things that AIs love: tools that are easy to use and honest about their results.
• Human outcomes. Behind every AI is a person trying to get something done. Getting bad data is worse than getting no data.
• Safety. Analog returns deterministic structured data, keeps extraction URLs out of its logs, and stays recognizable and blockable on the web.

Read the note behind Analog →