Skip to content

Quickstart

View as Markdown

Use your agent’s browser to open a page, then give that page to Analog to search and save. The example below uses the Codex GUI’s browser tool; the terminal setup covers macOS, Linux, and WSL. If your environment cannot supply a capture, use Analog’s built-in browser.

On this page

Copy this prompt into a coding agent that can run commands on your machine. It checks the existing setup, completes the steps that apply, and asks for your participation when needed.

You can read the agent setup instructions, or follow the terminal steps below. For an app that uses MCP, follow the MCP setup guide.

Give your agent the setup instructions

Set up Analog in this environment using
https://getanalog.io/agent-onboarding.md.
Check the existing setup and complete the steps that apply.
Use my agent's browser when it can supply a capture; otherwise use Analog's browser.
Verify it by finding Frankie on https://getanalog.io/about/ and searching the saved result again.
Tell me when a step needs my participation.

The installer adds the analog command to your PATH. Windows, Homebrew, pipx, and pip routes are in Install.

macOS, Linux, and WSL

curl -LsSf https://getanalog.io/install | sh

Already connected in this terminal? Skip to View a page.

Choose the command for your account. Both open your browser to verify your email. Return to the terminal or coding agent and wait for it to confirm the connection. There is nothing to copy into a config file.

Use this if you have an account, including one you just created on this website:

Connect an existing account

analog login

Use this instead if you do not have an account. Creating one during the private alpha requires an invite code:

Create a new account

analog signup

Open Analog’s About page in the Codex GUI’s browser. Follow that environment’s browser instructions to select the tab. This route needs supported JavaScript evaluation and Node filesystem access in the browser tool, with files accessible to the Analog CLI.

Prepare the helper once, then import it inside the browser tool. Capture writes the page to a file without changing it. Return only the receipt to the conversation; the HTML does not need to enter the model’s context.

Then ask view to find Frankie and save the result as about. Use the capture’s actual path and an unused saved name. Structured viewing sends the captured HTML and its URL to Analog; the result is saved locally. See Privacy for the data boundary.

Matching evidence appears on stdout. The receipt identifies the saved result by its handle; capture qualifications appear on stderr. The capture covers ordinary page content and visibility. Shadow content and frame documents are omitted, with observed omissions reported; see the capture contract.

Choose a new file path:

Prepare a reusable capture helper

analog browser capture-script --output ./analog_capture.mjs

The browser tooling supplies tab. Run this in its JavaScript runtime, using the helper’s printed absolute path and a new capture destination:

Capture inside the browser tool

const { capturePage } = await import("/absolute/path/analog_capture.mjs");
const receipt = await capturePage({
evaluate: expression => tab.playwright.evaluate(expression),
path: "/absolute/path/about.capture.json",
});
nodeRepl.write(receipt);

Find Frankie in the capture

analog view ./about.capture.json --save-as about --find "Frankie"

Search about again to find Marcus and Frankie together. This reads the saved result without fetching or extracting again. The name continues to refer to this save after you capture other pages.

An overview is optional: analog view about shows the page map, including navigation, records, and prose. You can also export or compare the saved data.

Ask two questions of the same save

analog view about --find "Marcus" --find "Frankie"

Each search retains its own matches. Read the source labels and qualifications with the evidence; a match in page prose is identified as such.

If your environment cannot supply a page file, use browse to acquire the URL with Analog’s built-in browser. You can also choose this route directly. It starts a fresh browser session; it does not use your browser’s sign-in. Analog downloads its browser when needed. Download size varies by platform and browser version.

Browse and search in one command

analog browse https://getanalog.io/about/ --find "Frankie"

Search that saved result again

analog view latest --find "Marcus"

Use the returned handle to identify this particular save; latest refers to your most recent save.

  • Working with results shows how to inspect, query, export, and compare saved results.
  • Fetching covers pagination, load-more buttons, and collapsed details.
  • View a page covers supplied HTML and the Python API.
  • MCP server connects Analog to Claude, Cursor, and other MCP-compatible agents.