Skip to main content
Once an agent is live, two screens matter: the Inbox for individual conversations, and Analytics for the shape of the whole thing.

​
The Inbox

Dashboard → Inbox, with separate WhatsApp and Instagram tabs. Every conversation your agents are having, in real time. For WhatsApp, pick the number whose conversations you want to see; search by phone number to jump to one contact.

​
Per-conversation controls

Clear Chat is about context, not display. It stops earlier messages being fed to the model — useful when a conversation has gone off the rails and you want the agent to start fresh with that customer.

​
Handed-off conversations

When an Instagram agent triggers human transfer, the conversation is marked as handed off and the bot stops replying to that contact. The Inbox has a filter for these, with a count — that filter is the queue your team works from. A handed-off contact stays handed off until someone clicks Assign to Agent. It never resumes on its own, so a customer waiting for a person is never interrupted by the bot mid-sentence.

​
Message delivery

Each outbound message carries its delivery state. A message that failed shows Not delivered with the provider’s reason, which is usually enough to tell a template problem from a number problem from a policy block.

​
Chat analytics

Dashboard → Analytics, chat section. Filter by:
  • Direction — all, outbound, inbound
  • Status — success (not failed), delivered (incl. read), read, sent (not delivered), failed
  • Channel — all, WhatsApp, Instagram
  • Campaign — all messages, campaign messages only, excluding campaigns
  • Agent — grouped into Live and In-Progress
Excluding campaigns is the filter that shows you what your chat agent is actually doing. Campaign sends otherwise dominate the totals and hide conversation volume entirely.

​
What to watch

The agent is talking more than it is being talked to. Check whether Multiple Messages per Reply is splitting too aggressively, or whether follow-ups are firing on conversations that were already finished.
Either the transfer criteria are too broad, or the agent genuinely cannot handle a category of question that deserves its own function.
Usually Context Length or Max Function Rounds was raised, or the agent moved to a larger model. All three are worth re-checking against the quality they actually bought.

Inbox

The Inbox in detail

Webhook

Getting these events into your own systems