SECTIONSAgent Chat Overview
Agent chat is the shared chat container that hosts the Veda Flex agents in the AI Connector package. Use it to ask an agent for help with a task in natural language.
Every Veda Flex agent formats dates and numbers to match your locale automatically, so responses use one consistent, unambiguous format throughout a conversation. Relative dates, such as "today" or "next Tuesday", are resolved using your time zone.
Agents
The agents available to you depend on the custom permissions your administrator has granted you. The following agents are available:
- Certinia Resourcing Agent Overview
- Certinia Services Delivery Agent Overview
- Certinia Customer Success Agent Overview
- Certinia Estimation Agent Overview
- Certinia Estimate Creation Agent Overview
If your administrator has not yet granted you access to any agents, the chat container opens with an empty state and you cannot send messages. Contact your administrator to request access.
Agent Selection
Agent chat selects an agent for you, or asks you to select one, based on the agents you can access.
- If you can access one agent, it is selected automatically when the chat container opens.
- If you can access more than one agent, a picker is displayed so that you can select one. The message input remains disabled until you select an agent.
After you select an agent, your choice is locked for that chat session. When you start a new chat session by clicking New Chat, your selected agent remains selected, and you can change it for the new conversation if you want. To work with a different agent, you can also close the chat container and open it again.
Chat Sessions
Each time you open an agent chat, a fresh chat session starts. You can also start a new chat session at any time by clicking New Chat, which clears the current conversation and keeps the selected agent. When you close the chat, your in-progress conversation ends, the agent selection clears, and the session is recorded for auditing.
You can refer back to something you discussed earlier in the same chat session, and the agent takes your earlier messages into account. A chat session has a maximum length, and when a conversation reaches it, the agent no longer takes your earliest messages into account.
For information about opening agent chat, see Opening an Agent Chat. For information about sending messages, see Sending Messages to an Agent.
Record Search and Results
You can search for a record by name or ID. For example, you can search for resources, assignments, resource requests, and projects by name or ID. Depending on the agent, you can also ask about skills, skill sets, and resource pools. If a search returns multiple results, the agent indicates the total number found. You can ask the agent to display results in a tabular format for easy comparison. The agent selects up to five of the most relevant data columns based on your query. If your results contain more than five columns, a message indicates how many are available, and you can ask the agent to show specific columns.
The agent displays up to 20 records at a time. If your query matches more records than that, the agent tells you which records it is showing you and how many records match in total. For example, it might report records 1 to 20 of 85. When further records remain, the agent tells you, and you can ask for them. For example, enter Show me the next set of results or Show me the rest. The agent displays one set of records for each message you send.
When you ask a question about the whole set of results, such as a count or a comparison, the agent answers from every record that matched your query, not only from the records it displayed. If the records are no longer available, the agent asks you to run your original request again. If a set of results is too large for the agent to work with at all, the agent asks you to narrow your query and run it again.
A table of records also includes a numbering column as its first column, headed #. The rows are numbered from 1, and the numbering continues across each set of records that the agent displays. The numbering column is separated from the data by a vertical line so that a row number is not confused with a value that begins with a number. This column is additional to the five data columns that the agent selects, and you cannot sort or resize it.
You can refer to a row by its number or by its position. For example, you can enter Hold #2, Assign the second one, or simply 2. The agent counts down the # column of the table that the agent displayed most recently, then acts on the record in that row. A table for a new query numbers its rows from one.
A number means a row number. Even if a record elsewhere in the table has an auto-number that ends in the same digits, the agent acts on the row, not on that record. To act on a particular record rather than a row, enter the record's full name or auto-number.
If the table contains no such row, the agent tells you which rows it does contain rather than acting on a different record. The exception is a bare number, such as 12, that matches the auto-number of exactly one record the agent has already shown you: in that case, the agent acts on that record. If the row number is greater than the records that the agent is showing you, the agent tells you the range of records that it displayed, and you can ask for the set that contains the row you want.
When the agent refers to an object record in its response, it displays the record name as a link you can click to open the record in a new browser tab. This applies to records in the agent's text responses, tables, and approval cards.
For information about querying skills, skill sets, and resource pools, see Querying Skills and Skill Sets Using the Certinia Customer Success Agent and Querying Resource Pools Using the Certinia Customer Success Agent.
Approval Cards
When an agent proposes to create, update, or delete a record, it displays an approval card. You can approve or reject the proposed changes.
The card opens with a short summary of the change. Depending on the number of records affected, the card displays either the record names or the total count. Record names are displayed as links that you can click to open the record in a new browser tab.
The summary is followed by a detail section that is collapsed by default. You can expand it and review the full set of values that the agent applies. The detail section lists a numbered block for each record in the change. Within each block, it lists every field that the agent sets.
If a change affects several records and the agent only updates some of them, the card displays the reason for each record that the agent didn't update. The changes that the agent applied are saved. If a change fails after you approve it, the card displays the reason. When the failure has a specific cause that you can act on, such as a validation rule on the record or a processing limit in your org, the card displays the cause for each of the records. When the cause is a connection, authentication, or configuration problem, the card displays a general message instead. If the change failed because it exceeded a processing limit, the agent explains what happened and asks whether you want to continue with fewer records or stop.
Navigating a Conversation with the Keyboard
You can read and work through a conversation using your keyboard. The conversation history is a single stop in the tab order, so pressing Tab moves focus to the history as a whole rather than to each message in turn. From there, you use the arrow keys to move between individual messages. Agent chat brings each message you move to into view.
The following table details the keys that move focus when the conversation history has focus.
| Action | Windows | Mac |
|---|---|---|
| Moves focus to the first message. | Home | Fn + Left Arrow |
| Moves focus to the most recent message. | End | Fn + Right Arrow |
| Moves focus to the next message. If no message has focus, moves focus to the first message. | Down Arrow | Down Arrow |
| Moves focus to the previous message. If no message has focus, moves focus to the most recent message. | Up Arrow | Up Arrow |
These keys apply only when focus is in the conversation history. When focus is in the input field, the up and down arrow keys step through the messages you sent earlier instead. For more information, see Sending Messages to an Agent.
When an agent asks you to confirm a change, focus moves to the Approve button on the confirmation card, so that you can respond without leaving the keyboard.
Response Feedback
You can rate the most recent response from an agent, so that Certinia can measure and improve the quality of agent responses.
The and buttons are displayed only while the agent's response is the last message in the conversation, and they are not displayed while the agent is still working. When you send your next message or attach a file, the buttons are withdrawn from the earlier response, and they are displayed again on the next response that the agent returns. The buttons are never displayed on your own messages, on attachments, or on approval cards.
A rating is final. When you click either button, that button is filled to confirm your rating, and you cannot change or remove the rating. Your rating is saved against the chat message, so it is still displayed if you refresh the page or resume the session. Ratings are never displayed as messages in the conversation. When you review a past conversation, only the final response can display a rating, and only if you rated it. Earlier responses display no buttons, whether or not you rated them, and you cannot rate a response in a past conversation.
Before you can rate a response, your administrator must grant you read access to the Response Feedback field on the Chat Message object. For more information, see Creating the AI Connector Permission Set.
Agent Chat Page View
You can continue an active conversation with an agent in the page view of agent chat. The Veda page gives you more space to work with detailed or complex responses, and a side panel where you can review your past conversations in read-only mode. For more information, see Veda Page Overview.