Observability ============= Once a Co-Worker is running, three tabs give live visibility into its behaviour and cost. They mirror the views in the :doc:`Command Center `, scoped to this Co-Worker. .. image:: ../images/coworker-analytics.png :alt: Analytics tab — metric cards, Agent Runs table, and breakdowns :align: center :width: 100% | .. list-table:: :header-rows: 1 :widths: 20 80 * - Tab - What it shows * - **Analytics** - A live dashboard: Total Cost, Total Tokens, Total Conversations, and Average Latency, plus input and output token counts. An **Agent Runs** table links to the traces for each run, and cost and token usage are broken down by user and by model. Filter by email, model, and time range; refresh, auto-refresh, or export to PDF. * - **Traces** - A paginated trace log. Open a trace to see the execution flow as a timeline of steps (spans), each with its input, output, metadata, duration, tokens, and cost. Skill invocations appear here as spans. Use **Add to Dataset** to capture a trace for :doc:`evaluation `. * - **Conversations** - A list of sessions with their status and duration. Open one to see the chat history, any KPI results, the session metadata, and a translate option. The Analytics dashboard is where per-Co-Worker **cost tracking** lives — use it to monitor spend by user and model before and after a wider rollout. Feedback from Users (Microsoft Teams) ------------------------------------- On Microsoft Teams, when a user reacts to one of the Co-Worker's messages — for example with a thumbs-up — the Co-Worker notices and uses it to adjust the tone of its next reply in the same conversation. The reaction shapes that one reply. This requires **Reflection** to be enabled on the :doc:`Memory ` tab (it is on by default). .. note:: **Coming soon.** How this feedback is surfaced in the interface is being finalised. Today the reaction influences the next reply; it is not yet shown as a separate score.