A PushMetrics agent at the center, surrounded by the things it can do: SQL, charts, email, Slack, Tableau and memory

An AI Agent is a chat-based assistant that lives inside PushMetrics. Ask it a question in plain English and it does the work: writes the SQL, builds the chart, drafts the email, posts to Slack, exports the Tableau view, runs the report. If you can do it by clicking around in PushMetrics, an agent can do it for you in a chat.

The thing that makes agents different from regular automations: there's no flow to build, no triggers to wire up. You say what you want. The agent figures out which tools to use and gets on with it.

What can agents do?

Query your databases
Ask a question, get a result table back. The agent writes the SQL, you don't.
Build charts
Turn the numbers into a Plotly chart you can hover, zoom, and share.
Send emails
Draft and send emails to the right people, with tables, charts, and CSV attachments.
Post to Slack
Drop updates, alerts, and reports straight into a channel or thread.
Export from Tableau
Find the workbook, apply filters, and export the view as PDF, image, CSV, or Excel.
Remember and learn
Keep notes between conversations, follow your team's playbooks, get sharper over time.

How it works

1. YOU ASK "Show me sales by region" 2. AGENT PLANS Picks tools and writes SQL 3. TOOLS RUN SQL runs, chart renders 4. RESULTS Table and chart appear in chat

The agent only ever uses the databases, integrations, and contacts you already have in PushMetrics. Workspace admins decide which of those each agent is allowed to touch.

You watch it happen. The SQL gets written in front of you, the chart appears as the data lands, and the message ships when the agent is happy with the draft.

Where you can talk to an agent

You don't have to be in PushMetrics to use one. The same agent works from a handful of places, and every conversation is saved, so you can pick it up later.

In the app
Head to the Requests page. Start a new conversation or open an old one from the list.
From Slack
@-mention the agent in a connected channel. The answer lands back in the same thread.
On a schedule
Have it run on its own. The classic example: weekly pipeline report, every Monday 9am.
Replay a past one
Re-run an old conversation in safe mode. The agent walks through the steps without actually sending the email or posting to Slack.

Stays inside the lines

Agents are powerful, which is why PushMetrics gives you a few knobs to keep them in check.

Per-conversation cap
If a single chat blows past your dollar limit, it stops. No runaway bills.
30-day budget per agent
Set a rolling 30-day cap. Hit it and the agent stops taking new requests until older spend ages out.
Per-tool permissions
You pick which tools each agent gets. If it's not switched on, the agent can't reach for it.
Full audit trail
Every message, tool call, and result lives with the conversation. Nothing happens off the record.

Getting started

🚀
Every workspace ships with a default agent called PushMetrics Assistant. It already has the usual tools and permissions wired up, so you can open the Requests page and just start typing.

Ready to build your own? Head to Configuring Agents.

A quick glossary

A handful of terms that come up everywhere in these docs:

Agent. An assistant you've named and set up. Each one has its own role description, its own model, and its own toolbox.
Conversation (or ticket). One chat with an agent. We call them tickets because each one has a status, an assignee, a priority, and a full history, the same as a help-desk ticket.
Tool. One specific action: running a query, sending an email, posting to Slack. Each call shows up as its own card in the chat.
Skill. A markdown playbook you write once and the agent follows. Think "how to send the weekly pipeline report": the same five steps, every time.
Memory. The agent's private notebook (one per agent), loaded into the system prompt at the start of every chat so the rules and context are present from message one.