AI Chat

The AI assistant is a first-class Chat hub in your dashboard — the same dual-sidebar layout as Analytics. Open it from the main sidebar, or press Cmd / Ctrl + Shift + K anywhere in the workspace for a compact slide-in panel. Start a new conversation, or pick up a recent one from the list on the left. Ask about campaigns, segments, funnels, and live workspace metrics; answers stream word by word as the assistant works. Replies can include charts and clickable chips that open the segment, journey, funnel, or template being discussed.

What you can do

• Open the Chat hub from the main dashboard sidebar (dual-sidebar layout like Analytics).
• Slide in a quick chat panel from anywhere with Cmd / Ctrl + Shift + K.
• Start a new chat with a centered prompt and one-click suggested prompts, browse all conversations on the Chats page, or continue any thread from Recents.
• Ask questions about your audience, journeys, funnels, popups, forms, and analytics — answers stream in real time and use the page you're on as context when helpful.
• Ask analytics questions in plain words: this period vs the last, trends over time, top pages, sources, campaigns, devices and countries, busiest hours, pages where people get stuck, who's online now, and how your SMS, email and web push messages were delivered.
• Look up a person by name, mobile number or email: did they come in today, what they usually do, or whether they did a specific action, with exact times and links to their profile and session replay.
• Find people who haven't been back for a while and see how many of them you can still reach by SMS or email.
• Ask the assistant to draft a new segment, message template, journey, popup, form, funnel, scoring model, or gamification campaign — you’ll get an approval card before anything is created.
• When the assistant finds live data, see charts and source chips under the reply, then open the item with one click.
• Switch Ask / Plan / Agent mode, use slash commands, or type @ to attach a segment, journey, funnel, popup, template, or form.
• Attach images with the + button, tap the mic to dictate — you'll see the sound bars move while it listens — and set workspace memory in Settings.
• Undo a recent create the assistant just made (within about 15 minutes).
• Copy, Retry, or Edit messages from the actions under a reply.
• Switch models from the prompt footer before you send the next message.
• Stop a reply mid-stream if you want to cut it short.
• See your wallet balance and this month's AI spend right above the prompt.

How it works

Navigation

The Chat hub uses the same shell as Analytics: a collapsed main sidebar (icon rail) that expands on hover, plus a dedicated chat column with New chat, Chats (full searchable list with timestamps), and a Recents shortcut list. Selecting any conversation opens the full thread in the main workspace.

Slide-in panel

From anywhere in the workspace, press Cmd / Ctrl + Shift + K to slide in a compact chat panel. It uses the same assistant as the hub. Open in Chat takes you to the full thread when you need more room. Press Escape or click outside to close it.

The panel and the hub both send the current page as context (for example the segment or funnel you're viewing), so follow-up questions can stay grounded in what you're looking at.

Conversations

Each conversation is stored under your workspace for you. The title is generated automatically from your first real question (truncated to a short label). A greeting like "hi" doesn't become the title. Hover a recent chat and open the ⋯ menu to star it (starred chats stay at the top), rename it, or delete it from Recents. If you keep Chat open in more than one browser tab, the recents list stays in sync when you create, rename, or delete a thread.

Streaming and tools

Replies stream as the model generates them. While the assistant is working, the chat shows a live thinking state that changes as it searches your workspace, drafts a create, or writes the reply. The prompt ring stays lit, and the send control has a metal rim so you can see the assistant is still on. When the assistant needs live data — user counts, segment sizes, recent events, workspace metrics — it uses read-only workspace tools. Results are woven into the answer; you do not need to approve read-only lookups.

When a visualization helps, the assistant can render a chart in the reply. When tool results point at specific segments, journeys, funnels, templates, popups, or forms, sources appear under the answer as clickable chips — expand Used N sources and open any item in the dashboard.

Write actions (creating or updating segments, templates, journeys, popups, forms, funnels, scoring models, and campaigns) are proposed with an approval card — nothing is created until you approve. Reject if it’s not what you wanted.

Journeys from chat follow a recipe: who enters (segment, schedule, event, or form), optional waits, then a send (SMS, email, web push, or popup). The assistant uses IDs it already found or created — it does not invent a canvas graph. New journeys stay Draft unless you ask to turn them on, and SMS/email only go live when a messenger is connected.

Analytics questions

The assistant reads the same analytics you see on the Analytics pages, so its numbers match the dashboard. Ask in your own words:

• How are we doing? Visitors, pageviews, sessions, bounce rate and session length, compared with the previous period.
• Over time. Daily or weekly trends for traffic or for any event you track: sign-ups, bookings, purchases, article reads, and so on. If an event carries a number such as an amount or a price, it can total or average it. Ask for a chart and it draws one.
• Breakdowns. Top pages, entry and exit pages, channels, referrers, campaigns, countries, devices, browsers, most frequent events, or the values of an event property.
• Timing and live activity. The busiest days and hours, and who's on the site right now.
• Problems. Pages with rage clicks, with links to session replays when they were recorded.
• Messaging. How your SMS, email and web push messages were delivered, plus bounces and unsubscribes.
• Audience. Total, identified and anonymous profiles, how many you can reach by phone, email or web push, and people who haven't been back for a while.

It doesn't assume you run a shop. It works out what matters from the events and pages in your workspace. If something isn't tracked, it tells you what to track instead of making up a number. Times use your browser's time zone, and when the dashboard is in Persian, dates use the Persian calendar.

Looking up a person

Ask about someone by name, mobile number or email: "Did Sara come in today?", "What does 0912… usually do?", "Did this customer ever submit the booking form?". The assistant finds the profile first. If several people match, it lists them and asks which one you mean. Then it answers with evidence: the exact time, what they did or which page they were on, a link to their profile, and a link to the session replay when one exists.

• Only identified people can be looked up: visitors who logged in or shared a name, phone number or email. Anonymous visitors can't be found by name.
• If the action you ask about isn't tracked on your site, the assistant says the records can't confirm or deny it. It never claims the person didn't do it. If it is tracked but there's no record, it says there's no record of it in that period.
• People lookups need permission to view contacts, and analytics answers need permission to view analytics.

Modes, slash, and @

Use the mode picker (or /ask, /plan, /agent) to choose how much the assistant can do: Ask stays in conversation context, Plan uses read-only tools, and Agent can also propose creates with approval. Type @ to attach a workspace entity as a chip for the next turn. See AI chat shortcuts for the full list.

Memory, attachments, voice, and undo

Add workspace memory under Settings — short brand or audience notes the assistant keeps in mind for every chat in that workspace. Attach images with + in the composer. Tap the mic to dictate; the prompt shows a live sound wave while it listens, then the words sit in the box. After an approved create, use Undo last create under the composer within about 15 minutes to remove that object.

Suggested prompts

The new chat screen shows a short row of one-click starter prompts under the greeting — workspace summary, list of segments, funnels, active journeys, or a quick chart if you have data. They send exactly the same text you'd type yourself.

Models

Use the model picker in the prompt footer to choose which model handles your next message. The list comes from your workspace's configured chat provider. You can use a fast economical model for quick questions and a stronger model for deeper analysis in the same thread.

Cost and wallet credit

The assistant is paid for from your workspace wallet, the same one you use for your plan. Each reply is charged by what it actually used, so a short question in a fast model costs a lot less than a long analysis in a stronger one. The charge shows up in Wallet history a few moments after the reply finishes.

The line above the prompt shows your current wallet balance and how much the assistant has cost this month. If the balance runs too low, chat pauses and you'll see a Top up wallet button. Teammates without billing access see a note to ask a workspace owner instead. As soon as the wallet is topped up, you can keep chatting.

Tips

• Ask a clear question ("what funnels do I have?", "summarise my last 7 days") — replies lead with the answer, a few useful points, and one next step.
• Start broad, then drill down in the same conversation.
• Recents shows your conversations for this workspace — each teammate has their own threads.
• In Persian (RTL), the sub-sidebar and prompt layout follow logical spacing so the chat column and input stay comfortable on the right.
• If Recents looks empty on a fresh install, run a workspace seed or send your first message — demo conversations may already exist in seeded workspaces.

Quick checks (good questions to try)

What's saved