Conversations inbox

Read every chat your agent has ever had — exactly the way the visitor saw it.

Sign in to your dashboard and open AI Agents → Conversations. You land on a list of every chat your agents have hosted, newest first. Click any row to read the full back-and-forth as chat bubbles, see what intent the agent picked up, and jump to the matching lead, booking or support ticket if one was created.

This chapter walks you through what the inbox shows, how to read a transcript, and the few questions it can answer that nothing else in the dashboard can.

#What this page is for

The Analytics page tells you how many conversations happened and how well they performed. The inbox tells you exactly what was said in any one of them.

You'll come here when:

• A customer calls and says "your bot told me X" — you need to verify what really happened.
• You want to read transcripts that didn't become a lead, so you can spot questions your agent isn't answering well.
• You're tuning your knowledge base and want to see which prompts the agent handled gracefully (and which it didn't).
• You need to export a transcript for a dispute, refund, or quality review.

If you only want to react to the next lead or booking, jump to the relevant inbox (Leads / Bookings / Support Tickets / Quote Requests) instead — those are action lists. Conversations is a read-only audit log.

#Finding the page

Sign in to your dashboard. In the left menu under AI Agents, click Conversations. The shortcut URL is:

/dashboard/{your-tenant-id}/agent-conversations

The menu item only shows up when you have at least one agent and your user has the View AI agents or Manage AI agents permission. If you don't see it, ask your workspace owner to grant one of those permissions in Team Management → Roles & permissions.

#What the list shows

Each row is one chat session. Columns, left to right:

You can hide or rearrange columns using the column toggle button at the top-right of the table.

#Tabs along the top

The inbox is split into six tabs so you don't have to filter from scratch every time:

Test traffic is hidden by default on every tab except Tests. This means your real numbers are never inflated by your own previews.

#Filters

The toolbar adds three quick filters on top of whichever tab is active:

• Status — Active / Closed.
• Intent — pick one of the canonical intents (Pricing, Booking, Support, Info, etc.) to narrow the list.
• Has lead / Has escalation — handy when you want both ticks at once (e.g. "show me escalations where the visitor also gave us contact info").

#Reading a transcript

Click View on any row to open the transcript page. You'll see:

1. Conversation metadata — agent name, when the chat started, the page URL, the referrer (if any), the intent, the status, the lead score, and the message count. Everything you saw in the list, plus the referrer.

2. Visitor contact — name, email, phone, company — but only when the visitor actually shared them. If the agent didn't capture contact info, this section is hidden so the page stays clean.

3. Summary — a short, auto-generated recap of the chat (when available). Useful when a transcript is long and you just need the gist.

4. Transcript — the full chat, rendered as chat bubbles:

   • Visitor messages — primary-colour bubbles, aligned to the right. What the person typed.
   • Agent replies — neutral / grey bubbles, aligned to the left. What the AI said back, rendered with the same formatting (links, lists, bold) the visitor saw.
   • Tool calls — amber bordered cards. These appear when the agent ran an internal capability mid-chat — for example, looking up your knowledge base, booking an appointment, or sending an email. The card names the tool so you can see why the agent paused.
   • Token labels — the small grey "N tokens" tag under each agent bubble tells you how much language-model usage the message cost. Helpful only when you're chasing a runaway bill.

The transcript is read-only. You cannot edit, redact, or delete individual messages from this screen — see 13a.7 below.

#Cross-references — jumping to leads, bookings, tickets

When a conversation produced a lead, booking, support ticket or quote request, the related screens link back here using the same conversation_id. From an entry in Leads / Bookings / Support Tickets / Quote Requests, click the conversation link in the detail panel to land on this transcript page.

The reverse — jumping from a transcript to its lead/booking — is on the roadmap (see gaps.md § 7.4.10 for the planned cross-links).

#Permissions

Permissions are managed in Team Management → Roles & permissions. The agent itself never sees who is reading transcripts; this is dashboard-only and never exposed to visitors.

#What this page cannot do

By design, the inbox is read-only. That keeps the audit trail trustworthy.

• ❌ You cannot edit a message. What the agent and the visitor said is what's stored.
• ❌ You cannot delete a conversation. If you need a transcript scrubbed for compliance (e.g. a GDPR erasure request), file a request with your workspace owner; we handle that out of band so the deletion is logged.
• ❌ You cannot reply to a visitor from here. The agent runs in real time on the visitor's page; once they leave, the chat is over. To follow up, use the contact info on the matching Lead entry.
• ❌ You cannot export to CSV from this page yet. Need bulk export? Use the REST API (Pro / Custom tiers only) to pull conversations on a schedule.

#Common workflows

"Did the agent actually quote the wrong price?"

1. Customer phones, references a chat from yesterday afternoon.
2. Open Conversations, filter Intent = Pricing, sort by Started desc.
3. Find the row that matches the customer (page URL or contact name on the lead).
4. Click View, scroll to the agent bubble that mentioned the price.
5. If the agent was wrong, screenshot the bubble and head to AI Agents → Knowledge documents to fix the underlying source.

"What questions are people asking that we're losing?"

1. Open Conversations, switch to the All tab.
2. Filter Lead = No and Intent = Pricing OR Booking.
3. Skim the first 10 transcripts. If you see the same unanswered question twice, head to AI Agents → Knowledge gaps — the agent has probably already logged it for you.

"My team needs to read every escalation each morning."

1. Bookmark /dashboard/{tenant}/agent-conversations?activeTab=escalated.
2. Open it first thing. Click through transcripts top-to-bottom.
3. Mark the matching Support Ticket or Lead as actioned in its own inbox.

#Privacy and retention

• Conversations live in your tenant's private database tables and are never shared with another tenant, even by accident — every query the inbox runs is scoped by your tenant ID.
• The transcript is what was sent to the language model: visitor messages are stored as typed, agent replies are stored as generated.
• If your knowledge base or system prompt references private business data, that content is only visible to the agent, not the visitor — and is never rendered in the transcript (you only see the visitor-facing reply).
• Retention is governed by your plan. The default is 18 months of full transcripts; older conversations are summarised down to metadata (no message bodies). Your workspace owner can request a longer retention on Custom plans.

#Related chapters

• Conversation analytics — the 30-day rollup view.
• Leads & hand-offs — the action list of conversations that captured contact info.
• Knowledge base — fix what visitors couldn't get answered.
• Managing agents — change the persona, tone, or escalation rules.

#Troubleshooting