AI agents overview

An AI agent is the personality answering on your channels. Every workspace starts with one — created during onboarding — and you can add more for different roles (pre-sales, order support, post-sales).

​

What an agent is

Each agent has:

​

How an agent actually replies

When a message arrives on a channel an agent is attached to:

1. Keloa picks the best knowledge passages for this question using semantic search.
2. The system prompt, the passages and the conversation so far go to the LLM (see Conversation memory).
3. The model drafts a reply. If it wants to use a tool (look up order, hand off), it calls it.
4. Reply goes back to the customer and into the inbox.

This all happens in a few seconds. The reply includes citations to the knowledge sources it used — you can see them in the conversation’s Sources panel.

​

Conversation memory

The agent doesn’t answer the latest message on its own. It reads the whole conversation the contact saw, on every channel:

• The most recent messages word for word, from the contact, the AI agent and your teammates: at least the last 6, and on WhatsApp, Telegram and Slack up to the last 20 (as long as they fit), so a burst of short messages stays intact.
• A summary of everything older. In a long conversation, Keloa summarises the earlier messages so the agent still knows what was already asked, answered or promised. The summary is updated as the conversation grows.
• Only what the contact saw. Internal notes, AI drafts waiting for review, discarded drafts and replies that failed to send are left out.

So a contact can say “and the blue one?” or “like I said earlier” and the agent understands. On Telegram and Slack, when a contact sends several messages in a row, the agent waits a few seconds and answers them together.

​

Earlier conversations

When Keloa can verify who is writing, the agent also gets a short summary of that contact’s earlier conversations, so it doesn’t ask again for something the contact already told you, like an order number.

• Only verified senders. The contact must be writing from the same Telegram account, WhatsApp number, Messenger or Instagram account, Slack account (in a direct message with the Keloa app; channel threads never use it), or from their own email address with a mail that passed sender checks (and nobody else on Cc). On the web widget, only a signed-in visitor counts. Anonymous website visitors never get it, even when they type a contact’s name or email address.
• Only conversations from that same sender. A conversation another person started, or an anonymous chat a teammate merged into the contact, is never used.
• Same way in. Only conversations on the same channel and the same bot, number, Page or Slack workspace, never another AI agent’s.
• Recent and short. Up to 3 conversations from the last 90 days, newest first, within your data retention period.
• Background only. The agent treats it as possibly outdated: the current conversation always wins, it never grants refunds or exceptions based on it, and it checks order details with its tools.

You can see which earlier conversations were used under Why this answer? on a reply. To turn it off for an AI agent, open AI agents → your agent → Advanced → Earlier conversations and switch off Remember the customer’s earlier conversations.

​

Grounded answers, not hallucinations

The agent only answers from your knowledge sources and the conversation itself. If it doesn’t know, it says so — or hands off to a human via a flow. You’ll never catch it inventing a policy you don’t have. If the agent is guessing when it shouldn’t, that’s almost always fixable by:

1. Adding the answer to your knowledge (a Q&A pair is fastest).
2. Tightening the system prompt — “If you don’t find it in the knowledge, say you’ll route to a teammate and hand off.”
3. Raising the confidence threshold so borderline replies escalate instead of sending.

​

Confidence threshold

Every reply gets a self-rated confidence score between 0 and 1. The agent’s confidence threshold (default 0.70) is the line below which the agent stops sending and escalates to a human via the configured handoff flow. Raise it to be stricter, lower it to be more aggressive. Confidence appears next to the AI badge on every assisted message in the inbox, and the distribution is plotted in Agent performance so you can see whether the threshold is calibrated.

​

Model providers

Keloa routes each agent to the AI model you selected. Model providers run with zero-retention settings, so prompts and replies are never used for training. See sub-processors.

​

One agent or several?

Start with one. Split when you notice:

• The system prompt is getting conflicting rules (“be playful on pre-sales, strict on refunds”).
• Different channels need different personalities (formal on email, casual on Instagram DM).
• You want different knowledge scopes (only product pages for pre-sales, only policy pages for returns).

Each agent can be attached to a different channel or different flow.

​

Next