Inbox — your unified inbox
Inbox is Repost21’s unified inbox. Customer messages from every connected channel land in one window: the AI agent replies on its own, and you only step in where a human is needed. It works in real time — new messages appear instantly, without refreshing the page.
Which channels land in the inbox
| Channel | How to connect | AI turns on |
|---|---|---|
| Telegram (bot) | Connect a bot in the “Social networks” section in “Inbox” mode | Automatically, right away |
| Telegram Business | Connect a bot to your Premium account (PRO plan and above) — the AI replies on your behalf | Automatically, right away |
| Connect the account and enable receiving messages | Manually, per conversation | |
| Facebook (Messenger) | Connect the page | Manually, per conversation |
| Connect a number (see Connections) | Manually, per conversation | |
| Web chat on the site | Create a webchat widget | Automatically, right away |
How the screen is laid out
The “Inbox” section has four parts:
- Agent rail on the left — your inbox agents that have conversations. Clicking one filters the list by that agent’s channels.
- Conversation list — search, the “Open / Needs human / Closed” filters, with unread items marked by a dot and a counter.
- Thread — the conversation with the customer, message search, and the ⋯ menu with actions.
- Contact card on the right — customer details, tags, and notes.
At the top there are the “Conversations”, “Live”, and “Web chat” tabs.
AI mode — the agent replies on its own
If an inbox agent is assigned to a channel (see AI agents) and AI mode is on in the conversation, the agent replies to the customer on its own. Auto-replies are labeled ”✨ AI” with the agent’s name — you can always tell where the AI wrote and where a human did.
In Telegram the AI reply “types” out gradually, like a live conversation partner — the customer sees the typing indicator and the message appearing.
Confidence indicator
Each conversation in the list is highlighted with a colored bar — this is the AI’s confidence in its last reply:
| Color | What it means | What to do |
|---|---|---|
| Green | The AI is confident — the reply is based on the knowledge base | No need to check |
| Yellow | Medium confidence | Review when you get a chance |
| Red | Low confidence or the AI handed the conversation to a human | Open first |
This way the operator only opens the problematic conversations instead of reading everything in a row.
Escalation to a human
When the agent can’t find an answer in the knowledge base, it doesn’t make things up: it politely tells the customer it’ll bring in a human and moves the conversation to the “Needs human” status. Such conversations are visible in the list filter and in Live mode.
Taking over a conversation
In the thread header there’s a “Take over” button — one click, and the AI goes quiet in this conversation, with the operator continuing from there. When you’re done — “Return to AI”, and the agent resumes replying on its own. Taking over affects only the current conversation; the AI handles the rest as before.
Voice messages
Voice messages from customers are transcribed to text automatically: the audio bubble appears right away, and below it the “Generating text…” indicator; a few seconds later the text is filled in. The AI agent replies to the transcription, just like to a regular message. Recognition works well for Russian, Uzbek, and mixed speech.
Translating incoming messages
Under an incoming message in another language there’s a “Translate” button — it translates the message into your interface language right in the thread. Clicking again brings back the original. Handy when customers write in a language the operator doesn’t speak.
AI reply language
The agent detects the customer’s language automatically from the first message of the conversation and replies in it — regardless of your interface language.
If you need to lock the language manually (for example, the customer switched languages mid-conversation), open the ⋯ menu in the thread header → “AI reply language” and pick the one you want. “Auto” returns to auto-detection and shows which language was recognized.
Working with conversations
- Statuses — open / closed; “Needs human” appears on escalation or takeover. The list filters are based on these statuses.
- Assigning an operator — via the thread’s ⋯ menu: the conversation is assigned to a team member (see Team).
- Snooze — hide a conversation until a chosen time, after which it returns to the open list on its own.
- Tags — add and remove labels on the contact right in the card on the right; tags are later used to segment broadcasts.
- Notes — internal comments about the customer, visible only to the team.
- Contact details — phone, email, channel, date of first contact and last activity. For WhatsApp the phone number is pulled in automatically.
Sending messages
In the thread composer you have:
- Text — with emoji support (a picker next to the input field).
- Photos and files — via the paperclip button: upload from disk, images are sent as photos, documents as files.
- Voice — record from the microphone right in the browser; in Telegram the customer gets a native voice message.
Live mode
The Live tab — a real-time picture of what’s happening, for a manager or agency:
- Event stream: incoming messages, AI replies, and operator replies — appearing instantly.
- Counters: open conversations, ones with an operator, unread, and ones with low AI confidence.
- Who’s working: AI agent activity over the day, operators online and their conversations, distribution across channels.
Handy to keep open on a second screen during peak hours.