I built a hosted AI agent that actually does things (not just talks)

Source: Dev.to

I’ve been frustrated with AI assistants that describe what they would do instead of actually doing it. So I built Cerebro.

What it can do

• Web browsing – A real headless browser that navigates to pages, clicks through links, and extracts what it needs.
  Example: “Go to Product Hunt and find today’s top 5 launches.”
• Web search – Real‑time search, not limited to training data.
• Email – Compose and send emails on your behalf.
  Example: “Research my top 3 competitors and email me a report.”
• Multi‑step tasks – Chain actions together (research + summarize + email) in a single command.

The uncensored part

Most AI assistants refuse or hedge on anything remotely controversial. Cerebro doesn’t. It presents all sides of any topic and lets you decide what to think, treating you like an adult.

Tech

Built on OpenClaw, an open‑source AI agent framework. The AI orchestrates tools in a loop: plan → execute a step → observe the result → continue.
The browser is a real Chromium instance accessed via the Chrome DevTools Protocol (CDP).

Every user session is isolated, and data never crosses between users.

Rough edges

• Complex multi‑step tasks sometimes need a retry.
• Heavy JavaScript sites can occasionally trip up the browser.

Still in beta.

Try it

Cerebro.vip – free, Google sign‑in only. No API keys, no setup.

I’m actively building based on feedback. Curious what the dev community thinks — especially what you’d break first.