Read dashboard metrics
Understand Talvi AI and MCP metrics, explore popular topics and Nuggets, and turn activity into improvements.
Use Dashboard to understand how people use Talvi, where they need support, and which knowledge and integrations deserve attention. Combine activity trends with topics, source usage, and feedback to decide what to improve next.
Choose a reporting period
Use Date range to choose a period, such as Last month, then select Talvi AI or, where enabled, Talvi AI MCP. The examples below use Last month. Your figures and configured limits will differ.
Cards labelled Selected range follow this selection. Request-capacity cards always describe today, the last hour, or the last minute; file inventory cards labelled All time describe the stored collection. A quiet minute can therefore appear alongside a busy month.
Explore Hot Talvi Topics
On Talvi AI, use the Dataset dropdown in the topic panel to switch between Hot Talvi Topics and Hot Talvi Nuggets. Here, Dataset chooses the keyword collection displayed in this panel.
| View | What it reveals | How to use it |
|---|---|---|
| Hot Talvi Topics | Keywords associated with what users are searching for. | Identify recurring questions, prioritise guidance updates, and check whether frequently requested subjects have clear, current sources. |
| Hot Talvi Nuggets | Keywords associated with generated Nuggets, highlighting potential knowledge gaps. | Find areas where a short learning intervention may help, then review and distribute suitable Nuggets. |
The chips show keyword occurrence counts. Selecting a chip highlights it; it does not filter every dashboard chart. Use Download to export the topic data. A frequent keyword indicates interest or a possible need for support, rather than proving a lack of competence.
For example, recurring payslip questions may suggest reviewing the payslip guidance and sharing a relevant Nugget. Follow distribute Talvi Nuggets to assign learning content, or analyse Nuggets with an AI report to explore actionable insights and recommendations.
Interpret the activity cards
Check request capacity
The three AI agent requests cards show usage against the configured daily, hourly, and minute capacity, with a percentage and usage bar. These count LLM executions, so they should not be read as a simple count of messages typed by people.
Use these cards when planning a rollout or investigating a burst of demand. A low daily percentage does not rule out reaching a shorter limit. If requests slow down or fail, follow rate limits and timeouts.
Understand adoption and agent usage
| Dashboard metric | What it shows | What you can learn |
|---|---|---|
| Talvi AI Queries (Selected range) | Requests to the Talvi LLM, shown across time. | See peaks and changes in demand after onboarding, communication, or a content update. |
| Active Talvi Chats (Selected range) | Chats with user activity in the period. | Understand ongoing conversation activity. |
| Total Talvi Chats (Selected range) | Chat sessions counted in the period. | Review conversation volume alongside active chats and users. |
| Active Users (Selected range) | Unique users in the period. | Understand how widely Talvi is being adopted. |
| Top 5 Active Talvi Users (Selected range) | Users with the most chat sessions. | Identify patterns of repeat use and opportunities to gather feedback. |
| AI Agent Usage (Selected range) | LLM executions grouped by AI Agent. | See which agents attract demand and where onboarding or content improvements may help. |
| Top Users Talvi AI Queries (Selected range) | Users and API callers by LLM execution activity. | Investigate concentrated usage, including automated activity. |
Some measures appear in more than one chart format. These are different views of the same activity, not separate totals to add together. Usage helps you understand adoption; pair it with User Feedback to assess whether answers are useful.
Review knowledge and learning activity
| Dashboard metric | What it shows | What you can do with it |
|---|---|---|
| File types viewed (Selected range) | Opened sources grouped by file type. | Understand how people use documents, media, and other supported formats. |
| Top Viewed Files (Selected range) | The most frequently opened source files. | Prioritise maintaining the guidance that people rely on most. |
| Talvi Nuggets (Selected range) | Nuggets generated during the period. | Review learning demand alongside Hot Talvi Nuggets and assigned content. |
| Data Imports (Selected range) | Files uploaded during the period. | Track updates to the knowledge base. Check import details when investigating processing issues. |
| File Import Type (All time) | Uploaded file types grouped by category. | Understand the mix of content in the collection. |
| Uploaded files (All time) | Total files and disk usage. | Monitor the size of the knowledge library and plan maintenance. |
For example, a heavily viewed document is a useful candidate for a regular accuracy review. Growing storage can prompt a Talvi Data Cleaner review for duplicate or outdated material. File views and Nugget generation indicate activity; they do not by themselves measure learning completion.
Optional MCP reporting
Select Talvi AI MCP to understand use of Talvi through connected assistants. This tab appears only where the deployment enables MCP. It reports Talvi MCP activity; it is not a combined report of every third-party tool configured in MCP Tool Store.
MCP capacity and participation
MCP requests today vs daily limit, MCP requests last hour vs hourly capacity, and MCP requests last minute vs minute limit show current request-counter usage against capacity. MCP Users (Selected range) shows unique users using Talvi MCP over the reporting period.
MCP calls and trends
MCP Tool Calls (Selected range) shows the total and its distribution over time. The MCP Users trend shows unique users by day. One assistant request can involve several tool calls, so tool calls and user questions are not interchangeable. Daily unique-user counts should not be added to calculate the period’s unique-user total.
Use these together to distinguish broader adoption from intensive use by a small group. A sudden spike can be a useful starting point for reviewing a new workflow or repeated requests.
Tools, agents, and outcomes
| Dashboard metric | What it helps you understand |
|---|---|
| Most Used MCP Tools | Which Talvi capabilities connected assistants call most often, such as search, answering, or opening sources. Use this to see how integrations are actually being used. |
| MCP AI Agent Usage | Which AI Agents receive MCP calls. Compare this with the agents and datasets you intended to make available. |
| Top MCP Users | Where MCP activity is concentrated. Use it to investigate unusual volumes or identify users who can offer workflow feedback. |
| MCP Success vs Failure | Recorded request outcomes. Investigate failures alongside the relevant logs, tool, and time. Successful execution does not itself prove answer quality. |
The outcome chart currently retains a “last 7 days” subtitle; its data follows the selected reporting range. Where failure data is available, MCP Failure Reasons groups errors by tool and reason to help narrow the investigation.
If expected activity is absent, check the reporting period and the connection’s user and agent in MCP access keys and authorisation. A missing MCP tab may simply reflect your deployment’s configuration.
Turn metrics into improvements
A short monthly review can connect the dashboard to practical action:
- Check Active Users, chat activity, and agent usage to understand adoption.
- Compare Hot Talvi Topics and Hot Talvi Nuggets to find recurring needs, then review source content or distribute relevant learning material.
- Review popular files and imports to keep the knowledge people use current.
- If MCP is enabled, review tool usage and outcomes to improve connected-assistant workflows.
- Check capacity during busy periods and use logs to investigate failures before they become recurring issues.
Use each card’s Download action for its CSV data when preparing a review. For individual events, open Logs Viewer; for answer quality, use User Feedback.
