Observation Dashboard
The Observation dashboard is where you monitor your agents in production. It shows conversation quality, volume and cost for a time window you choose, either across all agents or for one agent, and ends with a filterable conversation table that opens each session in Conversation Review.
Opening the Dashboard
Open the Observation tab of the AI Data Fusion web application. The dashboard opens with the scope and time range you used last (the first time: All Agents and Last 7 Days). Your last scope, time range and table filter are remembered in the browser.
The selected agent is also kept in the page URL as ?agent=<LU>/<agent>, so you can bookmark or share a link to one agent's dashboard, for example .../static/aifusion/observation?agent=customer_bank/banking. See Opening Observation for the full list of deep links.
Dashboard Layout
From top to bottom:
- Header bar — the title, the summary items for the period, the agent selector and the time range
- Number cards — single-value widgets (type
box) - Chart widgets — bar, stacked bar, pie, line and area charts
- Conversation Details — the conversation table, always last
Which cards and charts appear depends on the scope: the overall widgets for All Agents, or the selected agent's own widgets.
Header Bar
The header title shows the current scope: Agent Analytics | All Agents, or Agent Analytics |
Summary items
Next to the title, the header shows a row of summary items for the selected period. They come from the scope's summary widget — one item per row its query returns. For All Agents the shipped items are:
- Total — the number of analyzed conversations in the period
- Not Resolved — the percentage of analyzed conversations whose Resolution Status is Pending or Escalated
An agent's widgets.yaml can add its own headline items. For example, the banking sample agent adds Follow-ups (conversations tagged with the Follow-up signal) and At Risk (the percentage with a Medium or High Customer Lifetime Value Risk). A scope without a summary widget shows no summary items.
Agent selector
The agent selector lists All Agents followed by every deployed agent, shown as
- Replaces the widgets with that agent's own widgets (see Agent Scope below)
- Limits the conversation table to the agent's conversations, and hides the table's Agent column and Agent filter
- Updates the
?agent=parameter in the URL
If a remembered agent is no longer deployed, the dashboard falls back to All Agents.
Time range
The time range applies to all widgets and to the conversation table:
Time-series charts use 4-hour buckets for Last 24 Hours and daily buckets for the longer ranges.
All Agents Widgets
With All Agents selected, the dashboard shows the overall widgets defined in the widgets MTable of the aifusion LU (widgets.csv). The shipped set is:
There is no topics chart in the All Agents view: each agent has its own Topic values, so topics are shown per agent.
Token and step figures are filled in by the analysis process, so sessions that have not been analyzed yet do not contribute to them.
Agent Scope
When you select an agent, the dashboard shows only the widgets defined in that agent's Observation/widgets.yaml — its header summary, number cards and charts. The overall widgets are not shown. An agent without a widgets.yaml shows no widgets, just the conversation table.
Agent widgets read the agent's sessions only. If a widget's query is not written to filter on the agent's sessions, its card is marked Not filtered to this agent and shows data for all agents. See Configuring Observation for how to write agent widgets.
A new agent created in AI Fusion Studio comes with a starter widgets.yaml that includes the header summary, the four number cards, Top Conversation Topics and Token Consumption. The banking sample agent (customer_bank/AI/Agents/banking) adds customer-service views such as Urgency Level, Upsell Opportunity and Resolution Pathway.
Conversation Details Table
The Conversation Details table lists the sessions created in the selected period, newest first, for the current scope and filter. Click a row to open it in Conversation Review.
Columns
Topics, Resolution, Risk, Avg Response, Feedback and Summary are filled in by the analysis process. Feedback counters reflect the feedback that existed when the session was last analyzed.
Filters
Select a filter from the drop-down list next to the table title:
The High Risk Conversations filter relies on the Customer Lifetime Value Risk signal; it returns nothing for agents that do not define it.
Exporting the list
Click Export List to download the current list — the selected scope, time range and filter — as a CSV file. The file contains all the fields returned for each session, including fields that are not shown as columns. Use it for reporting or to load into an external analytics tool.
Next article: Conversation Review
Observation Dashboard
The Observation dashboard is where you monitor your agents in production. It shows conversation quality, volume and cost for a time window you choose, either across all agents or for one agent, and ends with a filterable conversation table that opens each session in Conversation Review.
Opening the Dashboard
Open the Observation tab of the AI Data Fusion web application. The dashboard opens with the scope and time range you used last (the first time: All Agents and Last 7 Days). Your last scope, time range and table filter are remembered in the browser.
The selected agent is also kept in the page URL as ?agent=<LU>/<agent>, so you can bookmark or share a link to one agent's dashboard, for example .../static/aifusion/observation?agent=customer_bank/banking. See Opening Observation for the full list of deep links.
Dashboard Layout
From top to bottom:
- Header bar — the title, the summary items for the period, the agent selector and the time range
- Number cards — single-value widgets (type
box) - Chart widgets — bar, stacked bar, pie, line and area charts
- Conversation Details — the conversation table, always last
Which cards and charts appear depends on the scope: the overall widgets for All Agents, or the selected agent's own widgets.
Header Bar
The header title shows the current scope: Agent Analytics | All Agents, or Agent Analytics |
Summary items
Next to the title, the header shows a row of summary items for the selected period. They come from the scope's summary widget — one item per row its query returns. For All Agents the shipped items are:
- Total — the number of analyzed conversations in the period
- Not Resolved — the percentage of analyzed conversations whose Resolution Status is Pending or Escalated
An agent's widgets.yaml can add its own headline items. For example, the banking sample agent adds Follow-ups (conversations tagged with the Follow-up signal) and At Risk (the percentage with a Medium or High Customer Lifetime Value Risk). A scope without a summary widget shows no summary items.
Agent selector
The agent selector lists All Agents followed by every deployed agent, shown as
- Replaces the widgets with that agent's own widgets (see Agent Scope below)
- Limits the conversation table to the agent's conversations, and hides the table's Agent column and Agent filter
- Updates the
?agent=parameter in the URL
If a remembered agent is no longer deployed, the dashboard falls back to All Agents.
Time range
The time range applies to all widgets and to the conversation table:
Time-series charts use 4-hour buckets for Last 24 Hours and daily buckets for the longer ranges.
All Agents Widgets
With All Agents selected, the dashboard shows the overall widgets defined in the widgets MTable of the aifusion LU (widgets.csv). The shipped set is:
There is no topics chart in the All Agents view: each agent has its own Topic values, so topics are shown per agent.
Token and step figures are filled in by the analysis process, so sessions that have not been analyzed yet do not contribute to them.
Agent Scope
When you select an agent, the dashboard shows only the widgets defined in that agent's Observation/widgets.yaml — its header summary, number cards and charts. The overall widgets are not shown. An agent without a widgets.yaml shows no widgets, just the conversation table.
Agent widgets read the agent's sessions only. If a widget's query is not written to filter on the agent's sessions, its card is marked Not filtered to this agent and shows data for all agents. See Configuring Observation for how to write agent widgets.
A new agent created in AI Fusion Studio comes with a starter widgets.yaml that includes the header summary, the four number cards, Top Conversation Topics and Token Consumption. The banking sample agent (customer_bank/AI/Agents/banking) adds customer-service views such as Urgency Level, Upsell Opportunity and Resolution Pathway.
Conversation Details Table
The Conversation Details table lists the sessions created in the selected period, newest first, for the current scope and filter. Click a row to open it in Conversation Review.
Columns
Topics, Resolution, Risk, Avg Response, Feedback and Summary are filled in by the analysis process. Feedback counters reflect the feedback that existed when the session was last analyzed.
Filters
Select a filter from the drop-down list next to the table title:
The High Risk Conversations filter relies on the Customer Lifetime Value Risk signal; it returns nothing for agents that do not define it.
Exporting the list
Click Export List to download the current list — the selected scope, time range and filter — as a CSV file. The file contains all the fields returned for each session, including fields that are not shown as columns. Use it for reporting or to load into an external analytics tool.
Next article: Conversation Review
