The Requests page in PushMetrics, with the conversation list on the left and an open conversation on the right

The Requests page is where you actually talk to agents. It's in the top nav, next to Home. Everything you've ever asked an agent lands here, so you can come back to old work, pick up a paused thread, or pass a result to someone else.

Starting a conversation

Once you're on the Requests page:

  • Type into the message box to kick off a fresh conversation.
  • Click anything in the sidebar to continue something you started earlier.
  • Flip My Tickets to filter out everyone else's chats and just see yours.
  • Past five conversations, a search box appears. It looks at titles and message bodies, so you can hunt down that thing you asked last week.

There's no save button. Everything's kept automatically.

What you see in the chat

The chat layout in PushMetrics: conversation list on the left, the message thread in the middle, and the message box at the bottom

The page is split three ways. Once the layout clicks, it's pretty self-explanatory.

The list on the left

Every conversation you can see, newest first. Each row tells you:

  • The title. The agent writes one off your first message; you can rename it any time.
  • A green dot if the agent is still working in that conversation.
  • A source icon: for the web app, for Slack.

The messages in the middle

The actual conversation. Four kinds of things show up here:

  • Your messages with your avatar.
  • The agent's replies, plain text mixed with code blocks, tables, and links wherever they help.
  • Tool cards that fire in real time as the agent runs a query, draws a chart, or sends an email.
  • Thinking steps. Collapsed by default; click to peek at how the agent worked through the problem.

The message box at the bottom

Type, hit Enter. The agent gets going immediately and the reply streams in word by word.

Tool cards

Every time the agent reaches for a tool, a card lands in the chat with what it just did. Nothing's hidden. The popular tools have their own custom card. Everything else (creating or querying a metrics view, recipient lookups, memory and skill edits, and so on) falls back to a generic wrench-icon card showing the call inputs and the result.

SQL
Syntax-highlighted query plus a sortable result table. Big result sets get a CSV download link.
List Tables / Describe Table
Lists tables in a schema, or shows columns and types for one. Mostly happens when the agent is feeling out a database for the first time.
Chart
Interactive Plotly chart. Hover for tooltips, drag to zoom. Falls back to a static image while it loads.
Email
Preview of the email with To, From, Cc, Bcc, subject, and body. Delivery status appears once it ships.
Slack
Preview of the Slack message and the destination channel. Status flips green or red after the post.
Tableau
Tableau searches show the query and what came back. Exports show the workbook, the view, and any filters applied.
Question
The agent's stuck and wants your input before continuing. Click a button or type a free-form reply. The card sticks around until you answer.
Tasks
A short to-do list the agent's working through. Checkmarks mark what's finished; an "In Progress" badge sits on whatever's happening right now.

Conversations as tickets

Under the hood, every conversation is also a ticket. That's how each one ends up with a few extra fields straight out of help-desk territory. It's also what lets you keep web chats, Slack threads, and scheduled runs in one place without losing track of which is which.

Title
A short summary written by the agent off your first message. Click to rename.
Status
Open while there's still something happening, Closed once you're finished with it.
Priority
Low, Medium, High, Urgent. Mostly useful when you're triaging a queue of requests.
Assigned To
Who owns the conversation. Usually the agent, sometimes a teammate you've handed it off to.
Requester
Whoever started the conversation: a teammate in PushMetrics, or the Slack user who @-mentioned the agent.
Tags
Click Add tags next to the title and start typing. Pick an existing tag or make a new one. Handy for grouping things together and filtering by topic later.
Source
Where the conversation came from: web app, Slack, scheduled run, or a replay of an earlier one.

Conversation status

Open or Closed, that's it. New ones start Open and stay that way until you decide you're done. Closing one clears it out of your active list, and you can flip it back to Open later if something else comes up.

Open
Something's still in flight, or you've parked it waiting for someone to act.
Closed
Done with it. Still fully readable, and one click brings it back to Open.

Chat stats

Right under the title is a small strip with the cost and usage for this conversation. Lets you keep an eye on spend without leaving the chat.

The chat stats strip showing total cost, input and output tokens, number of runs, and the model used

You'll see:

  • Cost so far in this conversation, in dollars.
  • Tokens in / out across all LLM calls.
  • Runs: how many times the agent went off and actually did work.
  • Model used in the most recent run.

Working with conversations

Once a conversation's running, here's what you can do:

  • Delete it. With a confirmation step, so you don't nuke something by accident.
  • Edit your messages. Change one of your past messages and the agent picks up from that point and answers again.
  • Share a single message. Grab a permalink to a specific message and send it to a teammate.

Live updates

āš”
Replies stream in live. Text comes in word by word, tools light up as they run, results land the moment they're ready. Close the tab mid-answer and the agent keeps going on the server. Come back later, the finished reply is waiting for you.