Skip to main content
Activity is the page in the Chatbase dashboard where you review everything your AI agent did. It holds the chat logs of every conversation on every channel (Chat Bubble, Help Page, Center Stage, WhatsApp, email, API, and more), the leads collected by the Collect Leads Action, and the data gathered by Collect Data Actions, all filterable and exportable. Teams use Activity to find wrong answers, correct them with Revise, and pause abusive conversations.

What you can do:

Read real conversations to see where your AI agent falls short, correct its answers on the spot, and export what it collected.
Activity shows data per conversation. Analytics aggregates the same activity into totals, topics, and sentiment.

​
Conversations

The chat logs give you a detailed view of all user interactions with your AI agent, so you can review individual conversations and evaluate its responses. Each log includes user messages, AI agent responses, and any triggered Actions, helping you identify issues, optimize responses, and improve the overall user experience.
The Conversations list in Activity

​
Filters

Conversation filters are organized into three categories to make it easier to find the conversations you’re looking for:

​
Conversation

Filter based on general conversation details:
  • Date range: Choose the time period you want to review.
  • Channel: Filter by the channel where the conversation took place.
  • Conversation state: Filter by the current state of the conversation, such as Ongoing, Ended, Taken over, or Paused.
  • User: Filter conversations by a specific user.
  • Contains voice sessions: Show only conversations that include voice sessions.

​
Content

Filter based on the content and quality of conversations:
  • Topics: Find conversations associated with specific topics.
  • Sentiment: Filter based on the detected sentiment of the conversation.
  • Feedback: Filter conversations based on the feedback they received.
  • Confidence score: Filter based on the AI agent’s confidence score.

​
Activity

Filter based on Actions performed during conversations:
  • Action used: Find conversations where a specific Action was used.
  • Action type: Filter by the type of Action performed.
  • Action result: Filter based on the result of an Action.
  • Procedure used: Find conversations where a specific Procedure was triggered.
  • Procedure result: Filter based on the result of a Procedure.

​
Conversation details

Open a conversation and select the Details tab to see its general details, including the contact, source, status, sentiment, message count, country, and conversation ID. For conversations started on your website, such as from the Chat Bubble, iframe, or Center Stage, Embedded on shows the page where the conversation took place, for example https://shop.example.com/pricing. Use it to tell which site or page a conversation came from when your AI agent is embedded on more than one. Query strings are not recorded, since they can contain sign-in tokens or personal details. Some browsers report only the domain.
If you directly embedded the iframe before support for page paths was added, add referrerpolicy="no-referrer-when-downgrade" to your <iframe> tag. Without it, Embedded on shows only the domain. Embeds that use the script need no change.
Embedded on is not shown for conversations started before this field was added.

​
Revise answer

This feature lets you adjust the AI agent’s response if it wasn’t accurate or satisfactory. When you click Revise Answer, a form appears showing the user’s original question, the AI agent’s response, and a field where you can enter the expected answer. Once the answer is updated, the question and answer are added automatically to the Q&A section of your sources.
The Revise Answer form
Matched Q&A: When a response matches a Q&A source, you can click the matched Q&A to open it directly, without searching for it in your sources.
Opening the matched Q&A source from a conversation

​
Pausing a conversation

Pausing a conversation from the chat logs
You can pause a conversation manually from the chat logs. Open the conversation and use the pause control. While a conversation is paused, the AI agent stops replying to that user until you resume it. This is useful when a user is abusing the AI agent. Manual pausing is supported on these channels:
  • Chat Bubble, iframe, Center Stage, and Help Page
  • API
  • WhatsApp, Messenger, and Instagram
  • Playground, Q&A preview, and Action preview
Conversations can also be paused automatically when Spam detection flags them under Build > Guardrails.

​
Confidence score

This indicates how confident the AI agent is in its response, based on the sources you’ve trained it on. Review responses with low confidence scores and revise them to improve their accuracy.

​
Exporting conversations

You can export the conversations in the chat logs directly from the dashboard using the Export button. The export can be downloaded as JSON, PDF, or CSV.
Exporting conversations as JSON, PDF, or CSV

​
Automatic conversation ending

Idle conversations are automatically ended after 3 days of inactivity. If a conversation hasn’t received a new message within this period, it is marked as ended automatically.

​
Ending and resuming conversations through the API

Use the Update conversation status endpoint in API v2 to decide yourself when a conversation ends or resumes. Set status to ended, paused, or ongoing.
  • Ending and pausing work for any conversation, whatever its channel.
  • Resuming works for API conversations only.
Workspaces billed per conversation can’t resume conversations.

​
Continued conversations

On these channels, a user can send a new message after their conversation has ended:
  • WhatsApp
  • Messenger
  • Instagram
  • Email
  • Slack
  • Zendesk
  • Salesforce
  • Zendesk Messaging
When that happens, a new conversation starts with context from the ended one, so the AI agent replies as if the chat never stopped. The user sees one continuous chat. On the dashboard, you see two separate conversations linked to each other:
  • Previous conversation opens the ended conversation so you can read its history.
  • Show summary shows a recap of the ended conversation.
To see every conversation with the same user, use the User filter.

​
Search conversations

You can search within conversation text and filter conversations by Actions and their results. Search is typo-tolerant, so searching for “refnud” still matches “refund”. Search covers conversations from up to 1 year ago.
Searching chat logs for a word, with matches highlighted in the results

​
Review conversations efficiently

You don’t need to read every conversation. Once a week, set Date range to the past week and apply these filters one at a time, so the conversations most likely to need a fix come first:
When a topic grows in Analytics, click it on the Topics page to open its conversations in Activity.

​
Fix a wrong answer with Revise

Revise turns a correction into a Q&A Data Source. Q&A is for exact answers: use it when a question has one right answer that shouldn’t vary, and the AI agent gives that answer the next time anyone asks. For example, a user asks “Do you ship to Canada?” and the AI agent says no, though you started shipping there last month.
1

Find the conversation

Search the conversation text, or use the filters above.
2

Revise the answer

Open the conversation, click Revise Answer on the wrong reply, and enter the expected answer, such as “Yes, we ship to Canada.” The pair is added to the Q&A section of your Data Sources with a Q/A badge. If the reply came from an existing Q&A, open it from Matched Q&A and edit it instead.
3

Add other wordings

In Build > Data Sources, open the new Q&A and add variations of the question, such as “Can I order from Canada?”.
4

Fix the source behind it

If the wrong answer came from outdated content, such as an old shipping page, update or remove that source too, so it doesn’t shape answers to related questions. To find it, ask the same question in the Playground and click Show sources below the chat.
5

Test it

Ask the same question in the Playground. The AI agent picks up new sources within a minute, with no retraining.
Q&A is only for exact answers to specific questions. When the AI agent is missing background knowledge, such as product details or policies, add it as a Data Source, such as a file or text snippet. When the problem is how the AI agent behaves, such as its tone or scope, change its instructions instead.

​
Leads

This section shows the collected leads along with their submission date. Custom fields configured on the Collect Leads Action each get their own column. You can filter leads by date and export them as CSV or PDF. Configure the leads form under Build > Actions by creating a Collect Leads Action.

​
Collected data

This section shows the submissions gathered by your Collect Data Actions, along with their submission date and an Action column showing which Action collected them. Each field configured on a Collect Data Action gets its own column. You can filter submissions by date, export them as CSV or PDF, and delete the ones you select.

​
Frequently asked questions

Conversations are kept indefinitely. They’re removed only when you delete them yourself or delete your Chatbase account.
Search in Activity covers conversations from up to 1 year ago. Filters such as Date range, Channel, and User narrow the list further.
Within a minute. Revise saves the correction as a Q&A in your Data Sources, and the AI agent picks up new sources without a retrain button. Test it in the Playground.
Conversations end automatically after 3 days without a new message. A user can also click End chat in the Chat Bubble, and you can end one through the API. On WhatsApp, email, and some other channels, a new message after the end starts a linked conversation. See Continued conversations.
Yes. Delete all conversations in the Danger zone of your AI agent’s Settings removes them permanently, and they can’t be recovered. You can also let website visitors delete their own conversations with the Delete conversations capability on the Chat Bubble.

​
Next steps

Analytics

See the same activity aggregated into totals, topics, and sentiment.

Best Practices

Turn what you find in the logs into better instructions and Data Sources.

Collect Leads

Configure which lead details your AI agent captures, and where they go.

Guardrails

Set rate limits and spam detection so abusive conversations pause themselves.