How the Inbox thinks
Open Inbox from the left sidebar. Every conversation from every connected channel — WhatsApp, Instagram, Facebook, Telegram — lands in one list, most recently active at the top.
Here’s the one idea that makes the rest click. A conversation isn’t “open” or “closed” the way a ticket is. What matters is who is in control right now, and there are only three answers:
- You — no automation is running, so you’re the one replying.
- The AI is handling it — a workflow is live and the agent is replying on its own.
- The AI is waiting on you — it paused on a judgement call and needs a steer before it continues.
The list shows you which of those three a conversation is in before you even click. The reply box at the bottom of a chat then changes shape to match. Learn to read that one signal and the Inbox runs itself.
Read the list at a glance
Each row carries everything you need to decide whether to open it. From left to right: an avatar, the contact’s name (or their number if you don’t have a name yet), the last message — prefixed with You: when your side sent it last — the time, and a small stack of status markers on the right.
Tuesday 4:30pm works! And the monthly fee?
Can you send the July schedule?
You: I’ve looped in Aaron on this — back soon.
Got it, see you Tuesday!
Two controls at the top shape what you see:
- All chat / Unread — flip the whole list to just the conversations with unread messages.
- Search by name or phone to jump straight to one person.
When the agent has paused for input, a small banner appears above the list — 1 conversation needs your input — and tapping it filters the list down to exactly those. A Filtered: needs input chip shows while that filter is on; clear it with the ×.
Triage: what to open first
The status markers on each row are a built-in priority order. Read them in this sequence and you’ll always touch the right conversation first.
The agent hit a judgement call and paused. Amber left-bar, amber robot, and it’s pinned to the top of the list. These are the ones to open first.
A workflow is running and the agent is replying on its own. Blue robot, nothing needed from you — but you can step in any time.
A violet count or dot means unread messages; an badge means a teammate tagged you in a note. No robot means no automation is on it.
The robot’s colour is the whole tell: amber means it’s stuck and waiting on you, blue means it’s happily working. No robot at all means nothing is automating this chat — it’s yours by default.
Inside a conversation
Click any row to open the full thread. Messages are laid out like any chat: the customer on one side, replies from you or the AI on the other. AI replies carry a small agent name label so you can always tell a machine reply from a human one.
Two things share the conversation view:
- On the far right, the Lead Info panel shows everything Exabloom knows about this person — fields, tags, notes, pipeline stage. It’s fully configurable; see Customise the Lead Info panel.
- The header has a Mark as Unread button (your manual “revisit later” flag) and shows badges like DND if the contact is on Do Not Disturb.
But the part that changes with the conversation’s state — and the part worth understanding deeply — is the reply box at the bottom. It has three faces, one for each of the three “who’s in control” answers:
Tell the agent what to do — it can reply, look up info, update the contact, etc.
When the AI is handling it
While a workflow is running, the reply box is replaced by a slim bar: a green dot, the agent’s name, and “is handling this conversation.” You don’t type here — the agent is replying for you, and it’ll keep going until the workflow ends or it needs a decision.
One button sits on that bar: Take Over. It’s the escape hatch — the moment you decide a chat needs a human, this is how you grab it. (We cover exactly what taking over does in a moment.)
When the AI asks for your input
For judgement calls — pricing, exceptions, anything sensitive — the agent doesn’t guess. It pauses and the reply box becomes an Awaiting Your Input panel: an amber header with the agent’s name, the question it’s stuck on, and often a few suggested answers as tappable chips. Tap a chip to drop it into the box, or type your own.
What you type goes to the agent, not the customer. By default — Instruct Agent, the blue button, also fired by pressing Enter — the agent takes your note (“offer the promo and a free trial”) and turns it into a proper, on-brand reply, then carries on with the workflow. You’re coaching, not writing the whole message.
There’s a second mode tucked behind the caret on that button. Open it and you can choose Send word-for-word to contact instead — your exact text goes straight to the customer, unedited, for when you want to say it precisely your way.
Prefer to just handle it yourself? The same panel has an outlined Take Over button — that’s the next section.
Taking over for good
Take Over is a deliberate, confirmed action — it appears on both the AI-handling bar and the Awaiting-Your-Input panel. It does one decisive thing: it stops the agent and the workflow run on this conversation and hands you the wheel.
This will stop Bright Minds Assistant and the running workflow (New-lead nurture). You will handle this conversation manually.
Confirm, and the reply box switches to the normal composer — you are now replying. The thread keeps a quiet Conversation Taken Over marker noting who stepped in.
Replying yourself
When no automation is running — or after you’ve taken over — you get the normal composer. It works like any chat box, with a few moves worth knowing.
- Reply vs Internal note tabs. A note is visible only to your team — never sent to the customer — and you can @mention a teammate in it, which is what lights up the badge in their list.
- Type / to drop in a quick reply — a saved message you reuse often. On WhatsApp, the template button sends an approved template. Both are covered in Quick replies & templates.
- Use Mark as Unread in the header to flag a conversation to return to.
Moves to copy
Four everyday situations, mapped to the exact buttons. Adapt to your own conversations.
Start every Inbox session by knocking out the conversations the AI is waiting on.
Let the AI keep handling a chat but steer its next reply in your words.
A delicate chat needs a human. Pull it out of the workflow and reply yourself.
Need a colleague’s eyes without messaging the customer? Leave a private note and tag them.
Good to know & pitfalls
- Read the robot’s colour, not just its presence. Amber = waiting on you (open it now); blue = working fine (leave it); no robot = no automation, it’s yours.
- Instruct Agent keeps the AI running; Take Over stops it. If you just need to unblock one question, instruct — don’t take over, or the workflow ends for that contact.
- Your instruction isn’t the customer’s message — by default the agent rewrites it. Use the caret’s Send word-for-word option when you want your exact text to reach the customer.
- There’s no “close” button. Conversations don’t get archived or resolved — they just go read/unread. Mark as Unread is your manual follow-up flag.
- Internal notes never reach the customer. They’re team-only; @mentioning a teammate notifies them with the mention badge.
- Watch the WhatsApp window. Once the 24-hour window closes you can only send approved templates until the customer writes back.
Need a hand?
Our Singapore-based team is one message away — happy to help you get set up.