An AI agent that browses the web for you. Powered by Gemini 3.5 Flash & Stagehand.
Access is invite-only. Contact your administrator to be added.
◎ Browser Agent
Welcome. Tell the agent what to do, and watch it work in the live browser on the right.
connecting…
Members
People who can sign in with their Google account.
What costs more, what costs less
Every task is billed by tokens — roughly, chunks of text or
image the model reads (input) and writes (output). Prices for Gemini 3.5 Flash:
Input (page text, screenshots)
$1.50
/ 1M tokens
Output (the agent's thinking & answers)
$9.00
/ 1M tokens
Cached input (repeated context)
$0.15
/ 1M tokens
▲ Costs more
Long, vague tasks. Every step is a fresh call that re-reads
the whole conversation. Steps are the #1 cost driver.
Multi-step flows. The agent sends a fresh screenshot
(~1–2K tokens) every step, so more steps means more cost.
Lots of agent reasoning. Output tokens cost 6× input,
so a chatty agent bills fast.
▼ Costs less
Precise instructions. "Get the top 3 titles from
news.ycombinator.com" beats "find me some news" — fewer steps, less wandering.
Simple / sparse pages. Less to read per step.
Longer runs benefit from caching. Repeated context is billed at
1/10th price, so later steps get cheaper.
Typical task: 1–3¢. A worst-case run that hits the
25-step limit tops out around 35–40¢. There are no spending limits — the
numbers below each answer show exactly where the money went.