Tutorial 29 · Telemetry: watching your agents work and knowing they are fine

What the video says

1. Telemetry is the control board for your agents: health, load and connections, in real time. It refreshes itself every eight seconds while you keep it open.
2. At the top, eight summary cards. The first says how many agents are within their working hours, out of the total. Then the jobs of the last twenty-four hours, and the failures, which turn red if there are any.
3. Then the tokens for the day against the running total, the positive and negative ratings you have given, the historical runs, and the turns carrying personal data that were handled locally, without going out to any provider.
4. The map draws the org chart live. Each circle is an agent and the lines run from boss to subordinate, exactly as you connected them in Autonomous Agents. You can drag the nodes to tidy the layout.
5. The colour of the ring is the health traffic light. Green is fine. Amber means it had a failure or is outside its hours. Red means more than a quarter of its jobs failed. And grey means it is switched off.
6. Click an agent and its record opens: its status, today's jobs and the total, the failures, the average duration, the tokens, which channel it receives on and which it delivers through, and when its last job and last heartbeat were.
7. The gap between last job and last heartbeat is the most useful thing here: if the heartbeat is recent but the last job was hours ago, the agent is alive with nothing to do, which is very different from being stuck.
8. Below, the same information as cards, one per agent, so you can compare them at a glance. Here you see straight away who is carrying the work and who is piling up failures.
9. The last twenty-four hours of activity: each bar is an hour, blue for jobs and red for failures. The last bar is the current hour. This is where you spot the peaks, and whether failures cluster at one particular time.
10. Happening right now is the live thread, with the time, who, and what. You will see an agent wake up, each step it takes, when it delivers and when it finishes. It is how you look over its shoulder without interrupting.
11. At the bottom, the recorded errors. Every failure is logged with the exact place in the code, what was being done, and a trace number. And the line above tells you which one repeats most, which is where to start.
12. Each entry unfolds: the request being served, the process where it broke, the technical error and the trace number to look it up in the logs. Personal data comes out masked.
13. The error list does not refresh on its own, so it doesn't shift while you are reading: press Refresh whenever you want it brought up to date.
14. How to use it day to day: look at the traffic light, and if something is red or amber, open that record to see whether it is failures or just working hours. If the failures repeat, the log below tells you exactly where.