Skip to content

Managing Conversations

The admin Conversations page (/admin/conversations) is where your team monitors everything happening in the organization’s chats: triage, intervention, and auditing.

Admin conversations screen

At the top, two tabs separate two very different worlds of conversation:

  • External Conversations — chats with people outside the organization: public chat (/a/:slug), embedded widget, WhatsApp, Telegram, and the API channel. This is where customers, leads, and public support land. This tab shows the state filters (Open/Human/Completed) and the Intervene button.
  • Internal Conversations — chats organization members had in the Hub (/hub/chat) with the platform’s agents. They serve as a history of internal agent usage. This tab has no human intervention — admins see the content in read-only mode.

The badge on the right side of the header shows quick counters: the globe counts external, the users icon counts internal.

Each item shows:

  • Avatar and name of the user (or “External Contact” / channel identifier when there is no profile).
  • Agent name answering the conversation.
  • Channel badge (TG, WA, API, Public, or Web) with its own color, plus a colored left border for emphasis.
  • Last message date/time.
  • Message counter (N msgs).
  • State badge: absent when the conversation is open (the default case, less noise), yellow for human intervention, green for completed.
  • Shield icon next to the name when AI is paused — same signal driving the Human state.

The list paginates infinitely — when you reach the bottom, the next page loads automatically.

Above the list you have several ways to narrow down:

  • Search by user or agent — free-text field filtering by user name, email, and agent name.
  • Channel — multi-select dropdown to include only WhatsApp, Telegram, API, public chat, or Web conversations.
  • Agent — dropdown to focus on one or more specific agents.
  • State (External tab only) — Open / Human / Completed pills with counters. See Conversation States.

Filters are reflected in the URL (?states=...&channels=...&agents=...), so you can save and share the link of a specific view.

When you open a conversation, the right panel shows two blocks:

  • Info — when the conversation started, last message timestamp, user data, the last agent that responded (important in squads with transfer), the last AI model used, and credits consumed.
  • Notes — structured memory the agent builds along the conversation via the Read Notes and Write Notes tools. See Internal Notes.

On mobile this panel becomes a side drawer (the “Notes” button in the conversation header).

  • Review real conversations before tweaking important prompts. The External tab shows how the agent actually replies to users, not what you imagine.
  • Use the Human filter as your inbox. Every conversation in that state is someone waiting for a human reply.
  • Complete what’s done. The Completed state keeps the Open tab focused on what needs attention.
  • Internal conversations become test cases. When an agent fails with a colleague in the Hub, the internal conversation is the best evidence for tweaking the prompt — without exposing customers.