fleet.seanwinslow.com Agent Fleet Observability
6/8 HEALTHY 2026-07-02 10:00 UTC
Vault Synthesizer · concepts written per night · 60 nights

4 nights clean

The synthesizer has written every night since the last dark night on 2026-06-28. The eval suite runs each morning and the fleet stays inside its governors. The worst regression it has caught is preserved on the chart below.

8-DAY SILENT REGRESSION 2026-05-04 2026-06-03 2026-07-02
Eval pass · last night
10 / 14
Fleet spend · 30 days
$15.64
93.8% local · 471 runs
Local-only share
93.8%
Most agents · $0/run
Spend governors
$50 / mo
$7 task · $20 daily breaker
healthy run failed run quiet day
vault_indexer
7 runs · 7d
vault_synthesizer
7 runs · 7d
deep_researcher
7 runs · 7d
meta_agent
quiet · 7d
daily_driver
6 runs · 7d
knowledge_lint
1 runs · 7d
flush
8 runs · 7d
job_feed
45 runs · 7d
vault_synthesizer
writes nightly concept summaries from the vault.
vault_indexer
maintains the embedding index for semantic search.
deep_researcher
answers ad-hoc research questions via cloud LLMs.
meta_agent
schedules and coordinates the other agents.
daily_driver
runs the morning briefing pipeline.
knowledge_lint
checks the vault for broken links and conflicts.
flush
clears agent caches and stale work queues.
job_feed
surfaces job leads from the watchlist (private surface only).
24-hour activity · 8 agents · 5 runs
vault_indexer
vault_synthesizer
deep_researcher
meta_agent
daily_driver
knowledge_lint
flush
job_feed
10:00 16:00 22:00 04:00 10:00
30-day cost trend · local vs cloud · stacked
2026-06-032026-07-02 · today
Model mix · last 30 days
93.8%LOCAL
  • local 93.8% 452 runs
  • cloud 6.2% 30 runs
Synthesizer telemetry · concepts written · 60 nights
2026-05-04 2026-06-03 2026-07-02
Eval suite · vault-synthesizer · latest run