Automa

Last reviewed Aug 26, 2026 · 859 words

The job that sent me to Automa was a supplier portal with no API, no export button and a login guarded by a session cookie that expired if you looked at it wrong. n8n could not touch it without a headless browser and an afternoon of selector maintenance. Automa did it in about 20 minutes as a drag-and-drop flow inside Chrome: open the page, click the month filter, read the table, write a CSV, scheduled for the first of the month. That is the entire niche: automation that must happen inside a real browser with your real logged-in session, built without code.

Automa lives in the browser, which is both the point and the ceiling

Automa is a browser extension for Chrome and Firefox, MIT-licensed, written in Vue, with 21,581 stars. Nothing runs on a server. Workflows are stored in the extension, execute in tabs the extension opens, and see exactly what you would see: the cookies, the logged-in state, the JavaScript-rendered content that a plain HTTP request never gets. That is why it handles sites that defeat every server-side tool.

The ceiling is the same fact. When the browser is closed, nothing runs. The catalogue's "browser extension only" con is the whole operational model, and the rest of this guide is about working around it.

What a flow looks like

The builder is a canvas of blocks joined by lines. Triggers include manual, interval, cron, a specific date, a keyboard shortcut, a context-menu entry and visiting a matching URL. Action blocks cover the things you do by hand: new tab, click element, get text, fill forms, take a screenshot, scroll, wait for an element, upload a file. Data blocks loop over a table of results, insert into Google Sheets, export to CSV or JSON, make an HTTP request, and run arbitrary JavaScript for anything the blocks cannot express. Conditions and delays give you branching and pacing. The element picker lets you click a thing on the page to capture its selector rather than writing CSS by hand, which is where most of the 20 minutes went for me.

Workflows export as JSON, so they live in your git repo and move between machines, and a community marketplace has ready-made flows for common sites.

Scheduling needs a browser that stays open

For a scheduled flow to fire, a browser with the extension loaded has to be running at that moment. On a laptop that means "when I happen to be logged in," which is fine for daily-ish jobs and useless for anything you rely on. The reliable pattern is a dedicated always-on machine: a mini PC or a VM on the homelab box, a separate browser profile logged into only the accounts the flows need, and the browser set to launch at login. Keep that profile out of your personal password manager sync, and treat the machine as holding live sessions to whatever the flows touch, because it does.

Chrome's extension platform limits how long background work can run, so long flows should be built to do their work in a visible tab rather than in the background, which Automa does anyway. Expect to re-check flows after a browser major version bump; the extension itself keeps up, but sites change their markup on their own schedule.

Pair it with n8n and let each do what it is good at

Automa has an HTTP request block and a webhook block, so the clean division is: Automa does the in-browser part and hands the result to n8n, or Huginn if you prefer it, over a webhook, and the server-side tool does everything else (parsing, deduplication, notifications, writing to a database). The reverse direction, n8n triggering Automa, is possible only while the browser is open, which pushes you back to the always-on profile. The Zapier alternatives page covers the server-side candidates; Automa is not one of them and is not trying to be.

Where it gets fiddly

The catalogue says complex flows get fiddly and it is right. Once a flow exceeds about 30 blocks with nested loops and conditions, the canvas is harder to read than the equivalent 40 lines of Playwright, and debugging means stepping through the log block by block. Sites with heavy anti-bot measures will detect scripted clicking regardless of the tool. Automa's sweet spot is flows you can hold in your head: log in, navigate, extract or submit, export.

What I'd do

Install Automa in a dedicated browser profile on the machine that already runs your homelab, log that profile into only what the flows need, and build one flow for the single most annoying site you deal with by hand. Send its output to n8n over a webhook and stop there; do not try to make Automa the hub. Keep each flow under 30 blocks, export the JSON into git, and accept that the browser has to be open, because no amount of engineering removes that and the tools that pretend otherwise are the ones that break silently.