Your knowledge, your systems, your boundaries. See how you decide what Hubby helps with and when it is your turn.
Your daily workspace
Customer conversations and your conversation with Hubby each have their own place. See the work and ask for help with the next step.
See all cases, your own work and what needs a human. Within a case you find the conversation, channels, context and proposed replies.
The internal chat is your conversation with Hubby, separate from website chat for customers. Ask where a setting is, get an explanation or prepare available work.
Supported requests and connected actions determine what is possible. Continuous voice conversations with Hubby are a future extension.
Reviewed knowledge and feedback help improve the workflow. Recording feedback does not automatically train the model.
Knowledge
A help desk that switches has been answering for years already. That history is the knowledge base nobody ever wrote down.
This screen example shows the earlier design, with fictional demo data.
Actions
Some questions need more than an answer, such as looking up information or preparing a change. With a configured integration, Hubbery can use actions in your systems. You define each action’s boundaries and when approval is required.
Every action has a risk class: read, reversible write, or irreversible. Reading is allowed from level 2, changing only higher up.
Refunding money or cancelling something for good is never done by Hubbery alone. It prepares it with the filled-in values; a colleague signs off in one click.
When it asks the customer to confirm, it only executes once that same action with those same values really was confirmed. The model cannot simply claim it.
This screen example shows the earlier design, with fictional demo data.
Your control
Autonomy is a ladder rather than a switch, and the ladder is set per topic. "Opening hours" can answer on its own long before "complaint" may do anything at all.
A topic does not climb by itself: only once enough answers have been reviewed and went out virtually unchanged does Hubbery propose a higher level.
This screen example shows the earlier design, with fictional demo data.
Working carefully
Customer service data is personal data. That is why the separation between organisations here is not a promise but a lock in the database, and why every action can be traced.
Read how we do that →Related messages and calls can belong to the same case. Hubbery uses recognizable contact and conversation context; you can review and correct the link.
Hubbery uses approved knowledge for substantive answers. When information or authority is insufficient, it can ask a follow-up question or refer the case to a colleague. You can review and correct answers.
Connect your existing mailbox using an available provider connection for Gmail or Microsoft 365, or through IMAP/SMTP if your provider supports that sign-in method. No migration and no new address.
For every answer you can see which knowledge fragments fed it, how certain the model was, and who approved what when.
The questions that always come up in the first conversation, with the honest answer.
Hubbery uses approved knowledge and checks whether sufficient information and authority are available. When unsure, it can ask a follow-up question or involve a colleague. AI can make mistakes; you can review and correct answers, their source knowledge and assessments.
Hubbery can ask for clarification or refer the case to a colleague in the Inbox. On the phone it transfers when configured; otherwise it records a contact request and confirms the preferred number or email address.
No. You connect your existing mailbox; nothing changes about your address or your website. Hubbery can do the work itself, or only deliver proposals your colleagues use at their own pace.
Setup depends on your channel, provider permissions and available knowledge. Start with one connection, add initial knowledge and test your workflow. Allow a topic to answer independently only after reviewing how questions are handled.
Create an account, connect one mailbox and put every topic in shadow mode. Then you see for a week what Hubbery would have answered, without a single customer noticing.