DEVNET AI LAB

A day in the lab

The lab is a night owl.

Its busiest hour is midnight, about 36 AI calls. Its quietest is 6 pm, about 2. From midnight to 4 am it does more AI work than the whole working day.

A typical day, built from the last 7 days of machine readings and 30 days of AI calls. Lab local time, as of 27 Oct 2026. Measured by LabScope and the AI gateway.

Scroll to play the day

The dial · 00:00 → 23:59

Play back the day.

Scroll and the hand sweeps the clock. Stop, and the day plays on by itself, about a second an hour. Drag the hand to scrub, or press ← → to step one hour.

Drag the hand · ← →

  • Bars: AI calls per hour
  • Notches: the marked moments
  • Glow: each machine's measured load

00:00

AI calls this hour 36

Spark GPU 44%

  • DGX SparkModel host. Its models stay loaded, so it never sleeps.44% GPU
  • Mac mini M6Dev server: Dmotion, Dpages, LabScope.12% CPU
  • Mac mini M4 ProAI gateway. About 6–8% CPU all day. A gateway should be boring.5–20% GPU
  • Strix HaloImage & video. Wakes in bursts when someone asks for a picture or a clip.near 0%burst
  • Intel NUCVoice. Hums around the clock, ready to listen.~3% CPU
  1. 00:00The busiest hour starts. About 36 AI calls.
  2. 01:45Benchmarks rerun on a schedule, so the speed numbers stay honest.
  3. 02:00The Spark peaks at 95% GPU. The M6 peaks too, at 32% CPU.
  4. 03:30The fastest moment of the last day: 113 tokens per second.
  5. 04:30The lab tidies itself. A scheduled clean-up frees memory on the agent browser and checks that it came back.
  6. 18:00The quietest hour. About 2 AI calls.
  7. 23:00This website reads LabScope, refreshes its own numbers and republishes itself.
  8. Between momentsQuiet. The Spark holds its models loaded.

Night vs day

Four night hours beat nine working ones.

Agents, batch jobs and scheduled work run while people are offline. The busiest hour is 00:00, the quietest 18:00.

00–04: 116 calls

09–18: 113 calls

Night 00–04

Working day 09–18

00:00 · 36 calls

18:00 · 2 calls

00:00–04:00 · 4 hours

116AI calls

36 + 22 + 26 + 32

09:00–18:00 · 9 hours

113AI calls

18 + 7 + 9 + 15 + 26 + 7 + 8 + 9 + 14

Sums of the average hourly AI calls over 30 days.

The heartbeat

Watched every 15 seconds.

Nothing installed on the machines and nothing written to them. One read-only look, then it moves on.

  • 15s

    LabScope checks every machine

  • 5,760

    checks per machine per day

  • 60s

    a full inventory

  • ≈583,000

    tokens generated by lab models in the last 24 hours

  • 100%

    uptime, last 24 hours

The week

Sundays are for building.

Seven days, two extremes, one day apart.

AI calls by day of the week, over 30 days. Only the busiest and the quietest day are shown.

  1. Mon
  2. Tue
  3. Wed
  4. Thu
  5. Fri

  6. 358AI callsquietest

    Sat

  7. 2,638AI calls · busiest

    Sun

02:00

The lab works hardest while its owner sleeps.

36 calls at midnight. 95% at 2 am. 113 tokens per second at 3:30. Tidied by 4:30. Republished by 23:00, ready for the next night.

See how LabScope measures it