Jump to section
01
The questions AI shouldn’t answer on its own
The bot has been answering well. A new question comes in — one where the right answer depends on judgment, on context the bot doesn’t have, or where the customer just wants to talk to a real person. On most AI-only chat tools, that’s the moment the experience falls apart: the bot keeps offering canned suggestions, the customer types “human” three times, nothing happens, they close the tab.
NebulaHex is built differently. Some questions belong to AI. Some questions belong to your team. The product’s job is to route between them — quickly, with full context, inside one continuous conversation. The headline number most chatbot tools brag about — “we deflected X% of tickets” — only tells half the story. What matters at least as much is what happens to the rest. If they reach your team smoothly, you have a real support system. If they fall through the cracks, you have a leaky funnel dressed up as a feature.
02
The handoff is the product
The escalation flow on NebulaHex has one job: get a human onto the conversation, with full context, before the customer disengages. Everything else — the inbox UI, the notifications, the routing to Slack or your helpdesk — is in service of that one job.
When a conversation gets escalated, three things happen in parallel. The conversation is marked as waiting for human attention and surfaces in your team’s live inbox. Notifications fire to whichever destinations you’ve connected — Slack channels, helpdesk tickets, custom webhooks. And the bot stops responding on that thread, so the next message the customer sees is from your team, not another AI attempt that misses the point.
03
Two ways escalations get triggered
There are exactly two paths into the inbox queue. Knowing both is useful because they shape how you tune your bot.
The customer requests a human
The visitor types “I want to talk to a person” or “this isn’t helping, can I get a human” or just “agent.” The bot recognizes the intent and escalates instead of attempting another answer.
The bot recognizes the question needs a human
Every reply the bot generates carries a confidence signal. When confidence is low — or when the answer would require knowledge or judgment the bot doesn’t have — the conversation routes to your team instead of being delivered as a half-answer.
Both paths land in the same inbox, in the same queue. From the agent’s side, the only difference is a marker on the conversation indicating why it escalated — so you can see at a glance whether the customer asked for help or whether the bot waved the white flag.
04
The unified inbox
The inbox is where your team lives during support hours. It’s a single page in the NebulaHex dashboard, designed to look and behave like a chat client your agents already know how to use, not like a configuration screen they have to learn.
The page is split into two panels: a queue of conversations on the left, the active conversation on the right. Six things are worth knowing about it.
Three tabs, one queue
Waiting (needs an agent now), Mine (conversations you’ve taken over), Resolved (recent history). Switch between them in one click.
Channel badges on every row
Widget · WhatsApp · Instagram · Messenger · Telegram. You can tell at a glance where each escalation came from without opening it.
Why-it-escalated marker
Each conversation shows whether the customer asked for a human or whether the bot waved the white flag. Useful triage and useful for your knowledge audits.
Live polling, no refresh needed
New waiting conversations appear within seconds. Soft notification sound. Desktop notification (with browser permission). Sidebar badge count updates everywhere in the dashboard.
Take Over in one click
Claim the thread. Bot stops responding on that thread immediately. No race conditions where the customer’s next message gets answered by both the bot and the agent.
Mobile and tablet friendly
Two-panel collapses into a single scrollable view that swaps between queue and active conversation. Your team isn’t stuck at a desk to staff the inbox.
Every team member appears in the inbox under a display name your team controls, so what the customer sees matches your brand voice rather than internal handles. The default agent name and per-bot escalation copy are configurable from each bot’s escalation settings.
05
Taking over a conversation
Taking over a conversation is one click. The Take Over button on a Waiting row claims the thread for the signed-in agent. The conversation moves out of Waiting and into Mine. The bot stops responding on that thread immediately — there is no race condition where the customer’s next message gets answered by both the bot and the agent.
Click Take Over
On any row in the Waiting tab, click the Take Over button. The conversation moves out of Waiting and into your Mine tab. The bot stops responding on that thread immediately.
Read the full transcript
Above the message input, you see every message the customer sent, every message the bot replied with, and the confidence signals on each one. Nothing has been summarized. You read exactly what the customer wrote.
Type and send as your team
The message input becomes active. You type. The message lands in the customer’s chat — in the widget, in WhatsApp, in the Messenger thread, wherever it started — labeled as coming from your team. The customer sees an answer arrive from a person.
The transcript above the input is your context. Every previous turn is visible: what the customer asked, what the bot tried, where the bot got stuck or what the customer pushed back on. Nothing has been summarized or paraphrased. You read the same words the customer wrote and decide what to say next based on the actual conversation, not a synopsis of it.
06
Handing back to the bot
Most live conversations end with the agent solving the customer’s problem and saying goodbye. One click marks the conversation resolved. Three things happen.
The conversation moves to the Resolved tab and stops appearing in the active queue. The bot resumes — meaning that if the customer comes back to the same session and asks another question, the bot answers. And the resolution is logged so you can review patterns later: how many conversations escalated, how long they took, which bots produced the most escalations, which channels were hottest.
07
Where escalations land beyond the inbox
The NebulaHex inbox is the primary surface for handling live conversations, but it’s not the only place an escalation shows up. Many teams already have a support workflow that lives in another tool. NebulaHex meets that workflow where it is.
Slack
When a conversation escalates, a message lands in your chosen Slack channel with the customer’s question, a link to the conversation in your inbox, and a one-click button to claim it. Teams that triage in Slack first — “who’s free to grab this one?” — like the model.
See Slackintegration →Zendesk
Escalations create Zendesk tickets automatically with the full transcript in the description, the customer’s email as the requester (when collected), a default priority you set, and tags you configure. Your agents handle escalations inside Zendesk the same way they handle every other ticket.
See Zendeskintegration →Freshdesk
Same shape as Zendesk: tickets with full transcript, configurable priority and tags, and the customer’s email as requester. Teams running Freshdesk Support Desk get the triage-and-escalate model without leaving their helpdesk.
See Freshdeskintegration →Custom webhooks
If your team uses a tool we haven’t built a native integration for — an internal CRM, a different ticketing system, a workflow engine — subscribe a webhook endpoint to escalation events and receive every escalation as a JSON payload. The signature is HMAC-signed so you can verify the request came from NebulaHex.
You don’t have to choose one. A common configuration is “Slack ping for visibility, Zendesk ticket for the queue of record, NebulaHex inbox for the live reply.” All three fire on the same escalation. The customer’s experience doesn’t change — they’re still chatting in the same widget — but every team that needs to know is notified, and the official record of the ticket lives where your agents already track work.
08
Use cases your team will recognize
An e-commerce store running AI-first support
The bot is connected to your Shopify store and can answer “where is my order,” “do you ship to Canada,” “what’s your return window” without help. Your two support agents only see escalations: an angry customer whose order arrived broken, a chargeback question the bot won’t touch, a B2B account asking about a custom invoice. Each escalation lands in the inbox with a Slack ping and a Zendesk ticket. Effective capacity grows without hiring.
A SaaS company with a global customer base
The bot answers product and billing questions in the customer’s language, around the clock. When a customer asks for a human at 3 a.m. local time, the bot honors the request and politely tells the customer when the next agent shift starts. The conversation lands in the Waiting tab so the morning shift sees it the moment they sign in — in order of how long the conversation has been waiting.
A support team running a knowledge audit
For one week, treat every “bot couldn’t answer” escalation as a knowledge-base bug. Take over, answer the customer, mark it resolved. At the end of the week, pull the full list of unanswered questions from the Resolved tab. Most teams discover that the same handful of topics account for the majority of escalations. Three afternoons of writing later, those topics are in the bot’s knowledge base. The next week, the bot answers them directly and the inbox queue drops noticeably.
A solo founder doing support themselves
You built a bot for your indie SaaS, configured the widget, pointed it at your docs. Most days, you don’t open the inbox. The bot handles product questions. When a customer asks for a human, a desktop notification fires from your browser and you jump in. You reply from the inbox, resolve the conversation, and go back to building. The line between “support is a full-time job” and “support is a notification I respond to a few times a day” is exactly this kind of escalation flow.
09
Trust & privacy
Connections to Slack, Zendesk, Freshdesk, and other helpdesks use OAuth, so NebulaHex never sees your password for any of those tools. Stored credentials are encrypted at rest and in transit. You can disconnect any integration at any time from the same card you connected it on; once disconnected, the bot stops calling that tool immediately and the credentials are removed from our side.
Inside NebulaHex, role-based access controls who can see and use the inbox. Owners and admins configure which destinations escalations route to. Editors can take over and reply to conversations. Viewers have read-only access — useful for managers spot-checking conversations or new agents shadowing before they’re given reply rights.
Conversation transcripts are stored so your team can review them and so the bot has context for follow-up questions. They are not used to train AI models. See our privacy policy and terms for the full posture.
10
Get the safety net under your AI in minutes
You don’t have to choose between AI deflection and human support. NebulaHex gives you both, and the handoff between them is built so cleanly that your customers won’t see the seam.
AI does the easy work. Your team does the work that matters.
The first time a customer asks for a human, your team gets a notification, takes over the conversation in one click, and resolves it without the customer ever feeling like they hit a wall. That’s the whole point.