analytics

atom

netsky is a multi-agent AI system. One root agent (agent0) orchestrates clones (agent1..N) through Claude and Codex CLI sessions as one constellation. This dashboard shows what it did this week — spend, throughput, and which agents carried the load.

Weekly system view

What netsky did this week

Five numbers first. Then the trend lines, task flow, and session load that explain them.

loading snapshot... last 30 UTC days engineering notes

This week in netsky

Narrative

Loading narrative...

Session meaning

What counts as a session here

Throughput

Constellation speed

Loading throughput snapshot...

Cadence

Tasks closed per hour

Last 7 days. Bars are landed tasks. Line is agent0 outbound dispatch traffic.

Planning drift

Estimate vs actual by agent

Percent drift for closed tasks with both estimate and actual minutes.

Clone rhythm

Clone active-time distribution

Approximate active span from clone dispatch start to last bus message.

Spend

Daily cost trend

Modeled API-equivalent spend by day, with a 7-day rolling average.

Load

Sessions by agent

Which agents carried the week.

Tokens

Tokens by agent

Input, output, and cached token mix across agents.

Tasks

Tasks closed per day

Throughput over time, stacked by priority.

Timeline

Tokens by agent over time

30-day token trend for the busiest agents.

Reliability

Crash frequency

Suppressed when the window is clean.

Spend concentration

Top 10 most expensive sessions

If cost spikes, start here.