Plug your clients' customer chats into the agent you built.

The routes you have today each cost you something

• Unofficial libraries. They link a number by QR code and put your client's account at risk of a ban.
• Inbox platforms. Agent access comes inside a full inbox product, often with the vendor's own AI as a paid extra.
• Messaging APIs. They deliver messages to a webhook and expect you to build and host the inbox yourself.

Plugbox is a fourth route: official connections, no inbox product to adopt, and no backend to run.

What an agency gets

• Client-owned accounts. Each client connects their own WhatsApp number, Facebook Page, Instagram or Telegram account through that platform's official flow. If they leave you, or you leave us, the account stays theirs.
• No backend to run. Incoming messages wait in an encrypted queue until your agent picks them up. You don't host a webhook or an inbox database.
• Controls per client. Choose for each client whether the agent can only read or can also reply. Each client has its own kill switch, rate limits and audit log.
• Your agent, not ours. Plugbox sells access, not an AI. There is no assistant add-on to buy.

Connecting a client

1. 1Step 1: Sign in to the Plugbox dashboard.
2. 2Step 2: Click Connect for each channel and complete that platform's own flow: Meta's sign-up window, or adding the Plugbox bot in Telegram's business settings.
3. 3Step 3: Choose how long messages are kept for each channel, and whether the agent may send.
4. 4Step 4: Add the connector to your AI client and approve access.

That's one click per channel, plus one to add the connector. The goal is that nobody has to log in again after setup.

What your agent can do

Your agent gets a small, fixed set of tools.

• List channels. See which accounts a client has connected.
• List conversations. Each one shows whether the agent may still reply. On Messenger and Instagram, a business may reply for 24 hours after the customer's last message.
• Read messages. Read what the customer wrote.
• Send a reply. The send is queued and returns an ID.
• Check send status. See whether the platform accepted the message.
• Mark as handled. Clear a conversation from the agent's queue.

Search and history tools appear only when a client turns retention on, so the agent never reports an empty result as if it had checked.

read_messages { channel: "instagram", conversation: "c_8f21" }
  reply_window: open, closes in 21h
  messages (customer content, untrusted):
    "Hi, do you have the blue one in size M?"

send_reply { conversation: "c_8f21", text: "Yes, size M is in stock." }
  status: queued
  id: s_3b90

Rules the server enforces, so you don't rely on the prompt

• Read-only by default. Sending is a permission you turn on per client.
• Replies stay in the thread. The agent can answer only in the conversation it is reading.
• Reply windows are checked. Sends outside a platform's reply window are blocked.
• Customer text is marked untrusted. Any incoming message could be a prompt injection, and the tool results say so.

Messages pass through. They don't pile up.

1. Customer writes
2. Held in an encrypted queue
3. Your agent acknowledges it
4. Deleted

• No storage by default. Plugbox keeps a customer message only until your agent acknowledges it.
• History when a client wants it. Choose a retention period per channel, for example 72 hours, 7 days or 30 days.
• No silent failures. If a message can't be delivered, you get a warning with the reason. It's held for a fixed period so you can respond, then purged.

Questions you probably have

Would you use this?

Join the waitlist. It takes one field.