Agents

Agents is where you see every governed agent you can reach and decide which one to ask. Each card answers two questions at once: what this agent is for, and whether it is sound right now.

What this screen is for

An agent is only useful if you know what it covers and can trust that it is running. This page puts both on the same card: the agent's mission and ready-made questions sit next to how its last run ended, its schedule and what it has published. You can go from "which agent knows about this?" to a chat with that agent in one click.

One time range sits above all the cards rather than inside each one, so the page stays calm whether you can reach three agents or thirty.

What you see

1. Agents — the page title, with The governed agents you can ask — start a chat from a card. underneath it.
2. Engine status line — one line for the whole page, with a symbol and a sentence: The engine is consuming work and running current code when all is well, The engine is not consuming work or Engine state could not be read otherwise. It is shown once, because it is a fact about the platform, not about any single agent.
3. Time range — four chips: Last 24 hours, Last 7 days, Last 30 days and All. The chosen range sets which output the cards count.
4. Agent cards — one per agent you can reach. Each card carries:
   • the agent's name, its type and its mission (hover the mission to read it in full);
   • a Follow button (it reads Following ✓ once you follow the agent);
   • how the last run ended — for example completed, cut by budget, errored, running or queued — and how long ago;
   • one sentence on what that means: The last run wrote 3 outputs., The last run turned up nothing new., The last run could not finish., Running now… or Has not run yet.;
   • the schedule in plain words, with its time zone, or No schedule set yet;
   • an output bar with a legend — Brief, Recommendation, Finding and No data in period, each with its number — or No output in this range;
   • severity badges (Critical, High, Medium, Low) with counts, when there are any;
   • when the last brief arrived (Last brief 2 hours ago) or why there is none;
   • ready-made questions built from the agent's data model, and an Ask button.

How to …

Start a chat with an agent

1. Find the agent's card.
2. Press Ask to open a new, empty chat with that agent.
3. Or press one of the ready-made questions: the chat opens with the question already typed in. Read it, adjust it if you like, and send it yourself.

Follow an agent

1. Press Follow on the agent's card.
2. The button changes to Following ✓.
3. From the agent's next run on, you receive a notification when it publishes new findings. Press the button again to stop following.

See what an agent produced in a given period

1. Choose a chip under Time range — for example Last 7 days.
2. Read each card's output bar and severity badges for that period.
3. Choose All to return to the full history. The choice lives in the page address, so it survives a reload, a bookmark and the back button.

How to read it

• The run line says how the last run ended; the sentence below says what it means. "Completed" can mean "wrote ten findings" or "looked and found nothing new" — the sentence tells you which.
• "Turned up nothing new" is a success. The agent looked at its data and there was nothing worth reporting. It carries a neutral symbol, not an error style.
• No data in period is a data state, not a finding. It counts periods where the agent looked and the data held no rows, and it never carries a severity.
• The schedule's time zone belongs to the agent. A daily 06:00 run happens at 06:00 in the zone shown next to it, not in yours.
• "No brief yet" and "does not include a brief" are different facts. The card tells you whether a brief is still to come or whether this agent's output contract simply does not publish one.

Good to know

• You see exactly the agents you have been given access to. If the page says You have no agents yet — an admin must grant you access., an administrator can grant access in the Central Management Console.
• Ready-made questions never send themselves. They only fill in the composer, so you always read a question before an agent works on it.
• Following is personal and only ever adds notifications for you; it changes nothing about the agent itself.
• Experiment runs never notify followers, so a trial run never reaches your inbox.

Related

• Chat — what happens after you press Ask
• Home — the day's work list built from these agents' output
• Timeline — every run of these agents, with what each one wrote
• What an agent can see — why each agent answers from its own data
• Settings — choose which notifications following an agent brings you