Someone is selling a “production-ready 4-in-1 FastAPI AI Automation Suite” for $8,000, and the listing has been circulating all week next to the n8n valuation chatter and the Relay.app team’s move over to Chrome. So I clicked through expecting junk.

It isn’t junk. FastAPI, Celery workers, Docker Compose, a document pipeline, a chatbot service, a scraper, an orchestrator to sit on top. Somebody actually built that, and eight grand for four working services isn’t a crazy ask if you were going to write them yourself anyway.

But every one of those four services is a server. And a server, sitting in a data center, knows absolutely nobody.

The credential tax nobody itemizes

Before that suite touches a single document on your behalf, you have to introduce it to every system it needs, in whatever dialect that vendor speaks: a Google Cloud project with a consent screen and Gmail scopes, a Slack app with a bot token, a HubSpot private key, a Notion internal integration, an S3 policy.

That’s the setup weekend. Then there’s the part after, which is worse, because keys expire and scopes drift and the person who created the Google Cloud project leaves the company, and now some Celery task is silently 401-ing at 4am while everyone assumes the automation is still running.

At most companies you don’t even own half those keys. Your admin does. So the real timeline on “production-ready” isn’t a weekend, it’s however long it takes three people to answer a ticket.

So what do people actually want automated?

Ask around and the answers are boring and consistent. Triage the inbox. Fill the same six fields in a vendor portal, forty times. Pull last quarter’s numbers out of a dashboard whose export button produces a PDF, of all damn things. Update the CRM after a call. Copy a list of 200 names from an internal admin tool into a spreadsheet.

Almost none of that has an API you can reach. Not because the API doesn’t exist somewhere in theory, but because it’s gated behind an enterprise tier, or a partner application, or a procurement portal built in 2015 that has a session cookie and no documentation. We wrote a whole post about what to do when there’s no API access, and the short version is that the login you already have is the integration.

Chrome did the auth work years ago

Your browser is holding live sessions for every one of those systems right now. Gmail, the CRM, the analytics dashboard, the internal admin tool nobody has touched since the last reorg. Cookies, SSO tokens, the MFA you cleared on Tuesday, all of it sitting in a profile on your machine.

Dassi runs in the side panel of that browser. It reads the page you’re looking at, clicks the buttons you’d click, fills the form you’d fill, and pulls the table you’d otherwise copy by hand. There is no integration to configure because there’s no integration involved. The agent is inside the session, not knocking on it from outside.

That’s a narrower tool than a four-service backend, and I’d rather be honest about the shape of it than pretend otherwise. It doesn’t run at 3am. It doesn’t scale to forty thousand invoices. It works when you’re at your desk, on the tab you’re already on.

Where the $8k version genuinely wins

High volume. Headless. Runs overnight with no human anywhere near it. Batch-processing invoices by the tens of thousands. That’s real engineering and it deserves a real server.

It’s just not what people mean when they say they want their work automated.

The model doesn’t have to come with the box

The other thing packaged suites do is decide your LLM for you. The provider is baked into the config, the token spend flows through whatever account the seller wired up, and switching later means touching every service.

Dassi takes the opposite position on purpose: log in with a ChatGPT subscription you already pay for, or drop in your own key for Claude, Gemini, DeepSeek, whatever you like this month. The prompts go from your browser to your provider, and nobody in the middle is billing you a margin on tokens or quietly keeping a copy. We’ve argued the BYOK case at more length here.

Cost of the browser agent, all in: zero dollars, zero servers, zero API keys, roughly ninety seconds of install.

I keep thinking about who buys the $8k suite. Probably someone who needs a backend and correctly decides not to write it. Fine. But I suspect a decent share of buyers are people who wanted their inbox sorted and their expense forms filled, and are about to spend a month wiring OAuth for work that a tab could’ve done before lunch.