In plain terms. It’s not a search box that runs one lookup. It reasons: it picks the right data, runs a query or two, reads the results, and sometimes asks a clarifying question before answering. Think analyst, not filter.
What you can ask
Ask about your on-chain audience, your email and campaign results, or both at once, which is where it shines. On-chain- Who holds your token or NFT, and how holdings are spread.
- A wallet’s balances, recent activity, and NFT holdings across chains.
- Token prices, and how a wallet’s portfolio value has moved.
- Who’s active on a contract, and who has approved it.
- Who opened, clicked, or bounced on a recent campaign.
- Which segment converts best.
- Look up a contact and see their engagement.
- “Which of my holders never open my emails?”
- “Top 100 wallets by portfolio value that went quiet this month.”
- “Holders of this collection who clicked the last campaign.”
It builds the audience for you
When an answer is a list of people worth messaging, ask the assistant to save it as a segment. It creates a real, reusable audience (named however you like) that shows up in the Segments tab and anywhere segments are used, ready to drop into a campaign or automation. You review it and launch; the assistant never sends anything on its own.It proposes, you approve
The assistant can go a step further and draft the next move, a campaign aimed at a segment, an in-app message, a retention play, but these are always proposal-first. It shows you exactly what it would do and waits for your go-ahead. Nothing sends, and anything it drafts stays a draft until you activate it. Building a segment is the one thing it does right away, because a segment only groups people, it doesn’t message anyone.Privacy holds
The assistant works with contact records and, at most, email domains, never raw email addresses. Segments store the people, not their addresses, and the send pipeline resolves each person to a channel only at send time. So you can analyze and build audiences freely without exposing anyone’s address. See Privacy and ZK.What it runs on
On-chain answers read from your enriched wallet data: each wallet’s portfolio value, activity, NFT holdings, and scores. The “Data enriched · N wallets” status at the top of Intelligence shows how many of your wallets are enriched and how fresh that is. Answers are only as complete as that set, so enrich more wallets or refresh before a big analysis. Two related notes:- Email-only contacts (no wallet on file) don’t appear in on-chain questions, but the assistant can still find them by their engagement.
- Results are bounded. Each step returns a capped number of rows and tells you when a list is cut short, so a “total” can be a large sample rather than every last wallet. Ask it to narrow the question when you need an exact count.
What it costs
Chat is the most capable AI feature, so it draws the most from your monthly AI-credit allowance: roughly 35 to 50 credits per turn, depending on how much reasoning the question needs (a quick lookup costs less than a deep, multi-step analysis). For comparison, generating a single SQL query costs 5.
When the monthly allowance runs out, chat pauses (the SQL editor keeps working, and running SQL is always free), and on pay-as-you-go it draws from your prepaid balance instead. Warnings appear as you approach the limit. See Plans and billing.
Getting good answers
- Name the specifics: the chain or contract, a time window (“this month”), and the goal (“for a win-back”). Vague questions get clarifying questions back.
- Ask for the audience you’ll act on, and save it as a segment in the same breath: “…and save those as a segment called Dormant Whales.”
- Spot-check before you send. Treat a chat-built segment like any draft audience: read a few rows against wallets you recognize before wiring it into a live campaign.