Your first conversation

You never start from a blank box. When your agent finishes booting, it sends the first message — introducing itself and suggesting where to begin. Your only job is to answer.

This page covers what to say back, and the handful of chat mechanics worth knowing on day one.

If you're not sure what to ask

Two places give you ready-made starting points:

• The Quick start rail on Home — a row of prompts right under the composer. Click one and it fills the chat for you.
• The Use cases catalog — 36 prompts across 6 categories, from portfolio questions to on-chain digging. Browse it from the sidebar, or read more in Use cases.

Both are just prefilled messages. Edit them, make them yours, and send.

Five good first prompts

Concrete beats clever. These five each exercise a different part of what your agent does:

1. Read your portfolio — "Give me an honest read of my portfolio: what's my biggest concentration risk right now?" (Works best after you connect your portfolio.)
2. Research something you hold — "Research my largest holding: what's driving the price this month, and what's the bear case?"
3. Set an alert in plain language — "Alert me if BTC moves more than 8% in a day." Your agent turns it into a durable rule; see Monitoring & alerts.
4. Teach it a preference — "Remember this: I invest on a 2–5 year horizon and I don't touch leverage." It saves that to memory and factors it into future answers.
5. Shape your briefs — "More macro context in my Morning Briefs, and always flag anything moving in my watchlist." Your agent remembers the preference and shapes future briefs around it.

What you'll see while it works

Longer questions take real work — reading data, browsing pages, running analysis. While that happens, the chat shows live status labels telling you what your agent is doing right now: thinking, browsing, analyzing, and so on. You're never staring at a silent spinner wondering if anything is happening.

Answers stream in as they're written, and streams survive page reloads — if you close the tab and come back, the response picks up where it was.

Redirecting a running answer

There's no stop button — a running response always completes. If it's heading somewhere you didn't intend, just send a follow-up message: your agent reads it as soon as the current answer finishes and adjusts course. "Actually, skip the price history — I only care about the tokenomics" works fine.

"Remember this"

Anything worth keeping, say so: "remember this" tells your agent to store it in its memory — preferences, context, decisions. For notes you both want to read and edit later, it uses the shared Notebook. It asks before saving; nothing lands in the Notebook without you knowing.

Attaching files

Drop files straight into the composer — up to 5 per message, 4 MB each. Supported: PDF, DOCX, XLSX, CSV, JSON, Markdown, plain text, and images (PNG, JPEG, GIF, WebP). You can also paste screenshots directly.

Good first uses: a screenshot of a chart you want a second opinion on, or a CSV of positions you'd like it to look at.

Asking about your portfolio

Once your wallets or exchanges are connected, every answer is grounded in what you actually hold — "why is my portfolio down today?" gets a real answer about your positions, not a generic market summary. Connecting takes a few minutes and is read-only, always: Connect your portfolio.

Next steps