How to Monitor and Review Your Agent's Conversations

Learn how to access the Sessions tab, how to search and filter sessions, how to review a conversation in detail, and how to use session data to improve your agent.

Once your AI Agent is live, you need to monitor how it handles real customer conversations. The Sessions tab gives you a full log of every conversation your agent has had — so you can review what happened, verify the agent behaved correctly, and identify areas for improvement.

How to Access Sessions

1. Go to the AI Agents page by clicking the AI Agent icon in the left navigation panel.

2. Click the Sessions tab at the top of the page.

---

What You See on the Sessions Page

The page is split into two panels. The left panel shows a list of all sessions. The right panel shows the detail of the session you select.

How to Read the Sessions List (Left Panel)

Each row in the list represents one conversation. Here is what each element shows:

Element	What It Shows
Bot name	Which AI Agent handled this conversation
Customer name	The name of the customer who chatted
Session ID	A short unique code for this conversation (shown in grey)
Channel tag	The channel — displayed as a "whatsapp" tag
Message count	Total number of messages in this session

---

How to Search for a Specific Session

1. Click the Search bar at the top of the sessions list.

2. Type the customer name or session ID.

3. The list filters as you type.

---

How to Filter Sessions by a Specific Agent

If you have multiple AI Agents and want to see sessions for just one of them:

1. Click the All Agents dropdown in the top right of the sessions view.

2. Select the agent you want to review.

3. The list will show only sessions handled by that agent.

---

How to Review a Session in Detail

1. Click any session row in the left panel.

2. The full conversation loads in the right panel.

3. White bubbles are the AI's responses. Green bubbles are the customer's messages.

---

What You See in the Session Detail View

Element	What It Means
Agent name (top left)	Which agent handled this chat
Customer and channel	Customer name and the channel (WhatsApp)
Timestamp	Date and time of the last message
Session ID	Full unique ID for the session
Refresh icon	Reloads the session to fetch the latest messages

---

How to See Which Tools Were Used

As you scroll through the conversation, you will notice tool indicators within the chat:

Indicator	What It Means
search_knowledge	The agent searched its knowledge base (FAQs or websites) to find an answer
Bot tool name	A Bot Tool was triggered — for example, "Talk To Human" or "Sell_Used_Car"
API tool name	An API Tool was called — for example, "Get Lead By Phone"

This tells you exactly how the agent arrived at each response — whether it used its FAQs, called an API, or triggered a journey.

---

How to View the Raw AI Context (LLM Context)

1. Locate any AI response in the conversation.

2. Click the LLM Context button below that response.

3. A panel opens showing the raw context that was sent to the AI model for that specific message.

This is an advanced feature — it shows you exactly what information the model had when generating its response. Useful for debugging when the agent gives an unexpected answer.

---

How to Use Sessions to Improve Your Agent

Monitoring sessions is not just about checking — it is about learning and improving. Here is what to look for:

Wrong answers. If the agent gives incorrect information, check your FAQs and Instructions. Either the FAQ is wrong, or a rule in your Instructions is conflicting.

Missed tool triggers. If the agent should have triggered a tool but did not, check the tool's Description. It may not be specific enough for the agent to recognise the intent.

Unnecessary escalations. If the agent hands off to a human too frequently, it may be missing FAQs. Check the Unresolved Intents tab and add the missing answers.

Slow responses. If response times are consistently high, consider lowering the Reasoning Effort in Settings or reducing the number of enabled tools.

High costs. If the per-message cost is too high, review whether the model, reasoning effort, and number of tool calls are all necessary for your use case.

How Often Should You Review Sessions?

In the first week after going live, review sessions daily. After that, review at least two to three times a week. Once your agent is stable and performing well, a weekly review is usually sufficient.

---

Summary

In this article, you learned how to access the Sessions tab, how to search and filter sessions, how to review a conversation in detail, how to check which tools were used, how to view the raw AI context, how to replay a session, and how to use session data to improve your agent over time.