The setup
- Multi-site Laravel production on Linux VPS
- PHP-FPM, Nginx, MySQL, Redis, queue workers
- Reflex agent + Brain in graduated auto-repair on validated playbooks
The problem
Like every small platform team, we had the usual patchwork: panel deploys, uptime checks, Slack alerts, and restart scripts that only worked when the author was awake. Revenue-impact incidents — 502s during checkout, queue backlog on mailers — were the ones that hurt.
What changed
With Reflex running, the failure modes that used to page a human now get detected, dry-run, and repaired automatically on the playbooks we trust — with a full audit trail behind every action. Fewer 3am pages, faster recovery, and one tool instead of three.
The hard numbers (platform telemetry)
We keep the named story qualitative on purpose. The numbers below are aggregated Reflex platform telemetry across live Linux servers — the single source of truth we cite everywhere.
- Incidents observed (platform)
- 4,312
- Resolved automatically
- 61%
- PHP-FPM repair speed
- 18× vs manual
- MTTR: manual → automated
- 43 → 4.8 min
Aggregated platform telemetry from the PHP production incidents 2026 study and ongoing Reflex Brain repairs. Not a guarantee for your fleet.
What Reflex caught
- PHP-FPM worker exhaustion / 502 recovery
- Queue worker stalls (Horizon)
- Disk trajectory alerts before full
- Post-deploy health regression (Pipeline gating)
“We built Reflex because we were tired of paging ourselves for the same PHP-FPM crash. Now most of those never reach a human.”
Read the parent-company story: How Net-3.co.uk Ltd runs its business on Reflex →