AgentQ Audit
The AgentQ audit is a workspace-wide record of everything AgentQ has done. It brings prompts, responses, usage, cost estimates, tool calls, and the platform actions AgentQ performed together in one view, so an administrator can see what was asked, what the model answered, and what changed as a result.
Only the Admin user role can open it, and it appears in the AgentQ sidebar only while AgentQ is connected. To work with it, see Select Audit Dates, Filter the Audit History, and Sort the Audit History.
What the Audit Covers
The unit of the audit is the turn, which is one prompt a user sent and everything AgentQ did in response. The audit holds every turn in the workspace, from every invoking user, in both active and archived conversations. Archiving a conversation keeps its turns here, but deleting one removes them, so a deleted conversation leaves no trace in the audit and its turns stop counting toward the totals. See Delete a Conversation. It opens on the last 30 calendar days and the period is adjustable. Longer result sets are paged, with controls above and below the history for moving between pages and for changing how many entries each one shows.
What an Audit Entry Contains
Each entry is one turn. Its header names the user who ran it and the time it was invoked, with a Provider badge for the provider that served it and a Cost badge for its estimate. Below the header sits the conversation the turn belongs to, marked Archived when that conversation has been archived, followed by four counts.
| Count | What it reports |
|---|---|
| Requests | How many calls the turn made to the AI provider. A turn that uses tools makes more than one. |
| Tokens | The total tokens the turn consumed, input and output together. |
| Tool calls | How many AgentQ tools the turn invoked. |
| Recorded actions | How many changes to platform records the turn produced. |
Two expandable sections hold the detail.
Content shows the Prompt exactly as it was sent and the Response rendered with its Markdown formatting. A turn where nothing was stored reads No response was recorded for this turn.
Tool calls and recorded actions lists each tool the turn invoked, with its name, its Arguments formatted as code, and its Output. Below the tools it lists the platform actions the turn produced, each naming the action and the record it affected, with the time and the transaction number it belongs to. The section appears only when the turn produced at least one tool call or action.
Cost Estimates
Every cost figure in the audit is an estimate Qualytics calculates from the token counts and the model's published pricing. It is not a bill from the provider and it does not replace one.
An entry carries a Cost badge reading Unavailable when no estimate could be calculated for the turn, which happens when the model has no pricing on record, when the turn predates cost tracking, or when no assistant response was stored. A missing estimate is a state of its own rather than a zero, so those turns are reported separately instead of being counted as free.
Selected-Period Summary
Beside the history, a summary totals the same turns the filters select. It covers the whole selection rather than the page you are looking at, so moving between pages never changes the figures, while narrowing the period or the users does.
| Metric | What it reports |
|---|---|
| AgentQ turns | How many turns fall in the selected period. |
| Invoking users | How many distinct users ran those turns. |
| LLM requests | How many calls those turns made to the AI provider. |
| Tool calls | How many AgentQ tools those turns invoked. |
| Tokens | The total tokens consumed, with the input and output split shown below the figure. |
| Estimated cost | The sum of the individual turn estimates. |
When any turn in the period has no cost estimate, a note above the metrics says how many, and those turns are left out of the total rather than counted as zero.
Changing a date or applying a user filter updates the entries and these totals together.
See Also
-
Supported AI Providers
Every provider you can connect, what the Beta badge means, and which ones take file attachments.
-
Amazon Bedrock Authentication
The three ways to authenticate to Bedrock, and what an IAM role setup expects.
-
Conversations, Responses & Context
How to write prompts, read AgentQ responses, and work with context-aware chats.
-
AgentQ Limits
Rate limits, token usage, timeouts, SQL constraints, and scope constraints.
-
Examples
Worked scenarios showing what you ask AgentQ, what it does, and the shape of the answer.
-
Best Practices
Prompt design, cost management, guardrail behavior, rate limits, and async operation patterns.
-
Permissions
The user roles behind chatting with AgentQ, configuring it, and reading the audit.
-
How It Works
Where AgentQ appears, how a turn runs, and what it is allowed to see.
-
The Chat Interface
Every control in the full-page and floating chat, and what the input accepts.
-
MCP
What the Model Context Protocol is, how it works, and why it matters.
-
AgentQ in Action
How Qualytics implements MCP, with its endpoint, tools, and tool step labels.
-
Tool Catalog
Every tool AgentQ can call, what each one does, and what it shares with the model.
-
Semantic Reporting
The structured query interface AgentQ uses to count, filter, and inspect platform resources.
-
Access Controls
How an administrator chooses what chat may share with the model, and what each level unlocks.