Product update · Conversations

Long conversations stay quick, and bot loops stop on their own

Long chats now stay fast because older messages are summarised for the AI, and a loop guard stops another bot from using up your responses on social channels.

Gopi Krishna Lakkepuram

· 3 min read

Shipped
Area
Conversations
Available on
Every plan

Most chats are a few messages long. Some aren't: a customer planning an event, comparing options over an afternoon, or coming back to the same WhatsApp thread all day. Until now, the AI reread the whole conversation before every reply, so very long chats slowly got slower. And on social channels there was a rarer problem: another business's bot messaging yours, each reply triggering the next, round and round. This update handles both.

What's new

Long conversations stay quick. Once a conversation gets long, the oldest messages are folded into a short running summary. Before each reply, the AI reads your chatbot's instructions, that summary and the most recent messages word for word. It keeps the thread of the conversation, such as the customer's name, what they asked for and what was agreed, without rereading every early message each time.

A loop guard on WhatsApp, Instagram and Messenger. Your chatbot now steps back when a conversation looks like a machine on the other end rather than a person:

  • Bursts get paused. If a contact sends messages faster than a person realistically types in a chat, the AI pauses replies to that contact. The conversation shows as Paused in your leads so you can look at it.
  • Runaway threads stop. A single conversation can't run forever. Past a high message count, the AI stops replying in that thread.
  • No reply without the responses to pay for it. On social channels, your chatbot now checks that you have AI responses left before it answers, instead of finding out afterwards.

In each case the incoming message is still saved, no reply is sent, and nothing is used from your plan.

Good to know

  • Nothing to set up. Both changes are on for every chatbot.
  • Your full history is kept. The summary is only what the AI reads. Studio still shows every message of every conversation, exactly as it was sent.
  • Summarising doesn't cost you anything. Writing the summary doesn't use any of your AI responses.
  • If summarising ever fails, the AI reads the full conversation instead. A reply is never held up or lost because of it.
  • Real customers aren't caught by the loop guard. It counts only the other side's messages and is set well above how fast people type in a chat. If it ever pauses a real customer, open the conversation in your leads and click Resume auto-response.
  • Website chat isn't affected by the loop guard. It applies to WhatsApp, Instagram and Messenger, where another business's bot can message yours.
For developers

Defaults on main: summarising starts once 20 messages have built up beyond the 10 most recent, which are always sent word for word. The loop guard pauses a contact after 20 inbound messages within 10 minutes and stops replying once a conversation reaches 150 messages. Channel conversations roll over after 24 hours, so the ceiling works out as a daily cap per contact. A pause uses the same per-contact auto-response pause as the Pause auto-response button.

Explore Hyperleap AI

An AI front desk that answers customer questions, collects contact details, and books appointments on your website and messaging channels.