> ## Documentation Index
> Fetch the complete documentation index at: https://docs.callkaro.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Conversation & Features

How much the agent remembers, how much it is allowed to say, and which WhatsApp features it uses.

<Note>
  This panel is for **WhatsApp** agents. Instagram agents have their own panel — see [Instagram Settings](/chat-agents/configuration/instagram-settings).
</Note>

## Context Length

`5` to `100` messages, default **10**.

How many previous messages are sent to the model as context. More context means better continuity in a long thread, but every message is tokens on every reply — so it is directly a cost and latency setting.

<Tip>
  10 covers most support and sales exchanges. Raise it when the agent visibly forgets something the customer said earlier in the same conversation; do not raise it "just in case".
</Tip>

Anything that must outlive the window — a customer's plan, their city, their language — belongs in the CRM contact record, which functions can read through `x_contact`.

## Max Bot Messages per Session

`0` to `200`, default **50**. A **session is a 4-hour window**.

The ceiling on how many messages the agent may send inside that window. `0` means unlimited.

<Warning>
  This is the backstop against a loop. A misconfigured agent that answers its own messages, or a customer that keeps triggering a send, is capped here rather than messaging someone hundreds of times. Setting it to `0` removes that protection entirely.
</Warning>

## Features

| Feature                          | Default | What it does                                                                                                                                                                   |
| -------------------------------- | ------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| **Audio Transcription**          | Off     | Transcribes voice notes the customer sends, so the agent can read and answer them. Without it, a voice note is something the agent cannot act on                               |
| **Support Interactive Messages** | Off     | Lets the agent use WhatsApp's interactive elements — quick-reply buttons and list messages — instead of plain text only                                                        |
| **Multiple Messages per Reply**  | **On**  | Splits a long reply into 2–3 shorter messages. Reads far more naturally on WhatsApp than one wall of text                                                                      |
| **Reply to Specific Message**    | Off     | The agent quotes the message it is answering, using WhatsApp's reply feature. Also enables per-question replies. Useful in busy threads where several questions arrive at once |
| **Image Context**                | Off     | The model can see and reason about images the customer sends — a photo of a damaged product, a screenshot of an error                                                          |
| **Gap Between Messages**         | 0.1s    | The delay between consecutive messages when the agent sends several. A slightly larger gap reads like a person typing                                                          |

<Tip>
  **Audio Transcription and Image Context are worth turning on for most support agents.** Customers send voice notes and photos without thinking about it; an agent that silently cannot process either looks broken.
</Tip>

## Related

<CardGroup cols={2}>
  <Card title="LLM Model" icon="microchip" href="/chat-agents/configuration/llm-model">
    Model, backup and function rounds
  </Card>

  <Card title="Follow-Ups" icon="clock" href="/chat-agents/configuration/follow-ups">
    What happens when the customer goes quiet
  </Card>
</CardGroup>