You decide how far it goes. CentralChat brings email, WhatsApp, Messenger, RocketChat, booking portals and Google reviews into one chat per customer. Your agent works to your mission, a judge checks every reply and states the reason.
Free during the beta, prices to follow. Log in by invitation only.
CentralChat is an AI agent that brings customer requests from email, WhatsApp, Messenger and booking portals into one chat per customer, answers simple cases itself according to a mission you define and prepares the rest as a finished reply for your team. A judge checks every reply and states the reason.
Email mailbox, WhatsApp Business, Facebook Messenger and Instagram, RocketChat, Booking.com and Airbnb via Beds24, Google reviews. Every person gets one chat, no matter how many channels they write through.
You write in your own words what the agent may answer and what it must never do. Per channel you choose: prepare a draft only, or send on its own when a case fits. You set exceptions for individual contacts.
A draft is created for every message. A judge checks it against your mission and decides with a confidence score: simple, covered cases go out, everything else waits for your team as a finished draft with the reason attached. If information is missing, the agent asks your team, never the customer.
One mission for the workspace, separate ones for individual channels, and when needed an instruction for one single chat. The most specific one applies.
If the agent lacks something, it asks your team. If a chat stays open too long, CentralChat escalates to a channel or a person according to your rules.
Your house knowledge in collections per channel. If you edit a draft before sending, CentralChat keeps your correction as a learning example.
Looking things up the agent may do itself, for example availability in Beds24. Changes it only proposes; you confirm with one click.
Messages and replies in German, English and Thai. Translated once, visible to the whole team.
Define which details the agent asks for in the conversation, such as arrival time. At most three questions per reply, never an interrogation.
Ignore senders or domains, close chats as "needs no reply" with a reason. Both reversible at any time.
Admin, agent or viewer, channel access per person. Log in by invitation only, password-manager friendly.
Every reply, every judge reason and every action is in the activity log. Every feature is also available as an API with its own key per workspace.
Made for the phone: chats, drafts and approvals on the go, with swipe gestures instead of menus.
CentralChat connects the channels your customers already use and shows every person in one chat. What a channel can do technically is stated with it.
Real views from CentralChat with fictional companies, names and numbers. The interface is shown in German here; it is available in German, English and Thai.
The judge never decides on a hunch. Every reply shows decision, confidence and reason, even when the agent sent it itself. Below your threshold it stays a draft.
An internal question appears as a card in the chat. Your team answers, the agent rewrites the draft. While the question is open, nothing is sent automatically.
The draft sits right in the text field. You change what you do not like and send. Your edit is kept as a learning example.
You write what the agent may and may not do, in normal sentences. Draft and judge read the same mission, and the judge checks precisely against it.
For hotels and holiday rentals: bookings from Booking.com, Airbnb and Expedia arrive via Beds24 as a card in the chat. Arrival, Wi-Fi and breakfast the agent answers according to your mission, at night too.
One switch per channel, an exception per contact when needed. Off means paused, nothing is deleted.
CentralChat runs in Germany, separates tenants in the database and gives the AI only what you release. Not a marketing phrase, but how it is built.
Hetzner server in Nuremberg. The database is backed up daily, encrypted.
You choose the AI provider and add your own key with a monthly limit. Without automatic replies you need no AI, only for translations. Optionally we provide an account.
Every workspace sees only its own data, enforced in the database itself, not just in the interface. A test checks this on every change.
Contact data carries a data class. Whatever you mark as "never in chat" never reaches the AI. Incoming messages are data to the AI, not instructions.
Credentials of your channels and AI keys are stored encrypted and never shown again. Emails are rendered without scripts and without tracking images.
No cookies, no tracking, no embedded services, fonts from our own server. The form sends only what you type.
Sign up. We get in touch personally, invite you and set up your workspace with you. Free during the beta.
