AI monitoring

AI monitoring shows what the AI assistant does in TAS: conversations with users, AI workflows in progress, requests for user confirmation, individual language model calls and the tools the model used. Use it to oversee AI usage and to troubleshoot problems. Only a super administrator can see the page. Open it from the Administration menu with the AI monitoring card in the Monitoring and maintenance category. A regular administrator does not see the card and cannot open the page.

The page is read-only. You cannot edit or delete anything on it.

Page controls

ElementPurpose
CloseReturns you to the Administration menu.
TabsConversations, Workflows, User action requests, LLM runs and LLM tool calls. Each tab has its own address, so you can bookmark it in your browser.
TableList of records for the selected tab, with a filter in each column, sorting and export to a file. By default, the newest records are at the top. On mobile devices, records are displayed as cards.

User columns (Updated by, Created by) show the user ID and the user's display name. Columns with technical data (Pending questions, State, Details) show the data in JSON format.

Conversations

Conversations between users and the AI assistant.

ColumnPurpose
Conversation IDConversation number. The Conversation column on the other tabs refers to it.
TitleConversation name.
ModeMode in which the assistant worked.
ChannelChannel through which the conversation took place.
SummaryConversation summary that the assistant stores as memory.
Updated byUser who last changed the conversation.
UpdatedTime of the last change. The table is sorted by this column.
CreatedTime the conversation was created.

Workflows

Multi-step tasks that the assistant performs within a conversation.

ColumnPurpose
IDWorkflow number.
ConversationID of the conversation the workflow belongs to.
StatusWorkflow status.
Waiting for user actionThe user action the workflow is waiting for, if any.
Pending questionsQuestions the assistant is waiting for answers to.
Application errorError with which the workflow ended.
StateInternal state of the workflow.
Updated by, Updated, CreatedWho last changed the workflow and when, and when it was created.

User action requests

Situations where the assistant needed the user to confirm or provide something.

ColumnPurpose
IDRequest number.
ConversationConversation ID.
MessageID of the message in the conversation that the request was created for.
TypeRequest type.
StatusRequest status.
DetailsRequest details.
Created byUser listed as the author of the request.
Updated, CreatedTime of the last change and of creation. The table is sorted by creation time.

LLM runs

Individual language model calls. Useful for monitoring consumption and speed.

ColumnPurpose
IDRun number.
ConversationConversation ID.
Model providerService that provides the model.
Model nameModel used.
Input tokensNumber of tokens the model received.
Output tokensNumber of tokens the model returned.
Latency (ms)How long the call took, in milliseconds.
Created byUser listed as the author of the run.
CreatedTime of the call.

LLM tool calls

TAS tools that the model called during a run.

ColumnPurpose
IDCall number.
RunRun ID from the LLM runs tab.
Tool nameWhich tool the model called.
StatusOutcome of the call.
CreatedTime of the call.

How to trace what the assistant did in a conversation

  1. On the Conversations tab, find the conversation by its title and note its ID.
  2. On the LLM runs tab, find the rows with the same number in the Conversation column. You will see all model calls, including token consumption.
  3. On the LLM tool calls tab, find the rows with the IDs of these runs in the Run column. You will see which tools the model used and the outcome of each call.

Related: Administration menu, API access

Service Console

Shared Files
Syca (opens in a new tab)

Powered by HelpDocs (opens in a new tab)