Six Ways Teams Use AgentZ¶

These six uses combine the same few parts. Those are an agent, a sandbox, a workflow or a dashboard, and an MCP connection for an outside tool.

HR Reviews Job Applications Every Week¶

HR receives applications by email with PDF resumes. A weekly workflow reads the Gmail mailbox in read-only mode and parses each resume. It scores each candidate against a rubric taken from the live careers page, then ranks them.

The workflow writes a versioned export and publishes six dashboards for leadership, recruiters and hiring managers. Edge cases take their own branch. If the careers page has no role text, the workflow uses the last valid rubric and logs it. If a resume fails to parse, it keeps the application and reads the email body.

The HR Leadership view dashboard with an Applications by role bar chart, a Fit mix donut chart and a Review funnel, under an agent picker, a dashboard picker and a date range picker — The Leadership view is one of the six HR dashboards.

The schedule runs every Monday at 03:30 UTC. Build it yourself with HR application intake.

A Webhook Starts a KubeArmor Alert Investigation¶

KubeArmor is a security tool that raises alerts. A KubeArmor alert calls a webhook. The webhook starts a workflow that investigates the alert and writes a report with a severity. Read Webhooks to set up the call.

An Alert Source That Sends HTTP Calls Can Start a Workflow¶

An alert source that can send an HTTP call can start a workflow the same way. An MCP connection gives the agent access to your ticket tool. The sandbox exposes only the tools the workflow needs. Alert text comes from outside, so treat it as untrusted input. See Zero Secret Exposure.

A Health Triage Demo Builds a Tamil Dashboard¶

The rural-health-triage workflow runs on Sarvam AI, a model provider for Indian languages. It sorts each complaint as urgent, routine or pharmacy-only. Prompts written to force a high priority go to human review.

A black banner with the AgentZ and Sarvam logos and the text AgentZ now supports Sarvam AI out of the box — The announcement banner for Sarvam AI support in AgentZ.

The same agent built a dashboard called PHC Triage Metrics, in Tamil.

The PHC Triage Metrics dashboard with a Triage History chart, a Recent Cases table with Tamil complaint text and a Triage Count gauge — The Recent Cases table shows complaints in Tamil.

Pick Sarvam AI in the Provider list. See Inference providers.

Agents Build Their Own Dashboards¶

Ask an agent for a view of your data. It picks the charts, builds the dashboard and keeps it fed. Twelve widget kinds exist, among them line, area, bar, pie, funnel, sankey, scatter, gauge and table.

The HR example above publishes six dashboards this way. A widget is one chart or table on a dashboard. Read Dashboards.

An example dashboard with a stepped line chart of open incidents, a donut chart of revenue by channel, a bar chart of pipeline by region and a bar chart of support load — An example dashboard that an agent built. It mixes line, donut and bar charts.

Your Own Scripts Become Tools¶

Upload a Bash, Python or Node.js script of up to 64 KiB. Describe when the agent should use it and list its inputs. The script reads its inputs as JSON on stdin and prints its result to stdout.

A call to the tool appears in Lens with its arguments and result. Follow Add a custom tool.

Next Step¶

Build your first agent with Introduction to the quickstart.