Your first conversation

What to expect the first time you ask your agent to do something real.

Chat looks like an ordinary messaging app, but the agent is not only talking — it can act. Ask it to check something on your machine and you will watch it work.

A real first conversation: the question 'What files are on my Desktop?', a collapsed thought-process line, two tool cards for a file search and a terminal command, and the agent's formatted answer. — An unedited first conversation. The agent tried a search, decided it was the wrong tool, ran a terminal listing instead, and explained what it found — every step visible.

What you see while it works

Thinking — a collapsed line you can open to read its reasoning. When it finishes, the line shows how long it took.
Tool calls — a card for each action: the tool, what it ran, and what came back underneath. A file edit shows the diff; a command shows its output.
The answer — streamed a word at a time as the model produces it.

Tool output stays folded so the conversation reads like a conversation — the card just says how many lines came back. Click it when you want the evidence:

The same conversation with the terminal tool card expanded, showing the full raw output of the ls command the agent ran. — The expanded card is the actual command output, not a summary — what the agent saw is what you see.

Things worth trying

“What files are on my desktop?” — watch it run a command and read the result.
“Remember that I prefer short answers.” — it writes that to memory and applies it from then on.
“Summarise this page: <url>” — it browses and reports back.

It asks before it acts

Depending on the safety mode you chose during setup, the agent will ask permission before anything that changes your machine. You can change that in Config at any time.