The AI SRE that fixes prod, not just describes it.
It watches your legacy monolith, proves each hypothesis with executable checks against the live system, and only then touches anything.
FineractApiHighErrorRate
5xx 4.1% over 5m · p99 3.2s
LockWaitAboveBudget
p99 2.4s on m_appuser
H3 lock tables held
confirmed against live state
Guard patch on 083
applied · rollback armed
Error rate 4.1% → 0.0%
trial balance intact
TokenRefreshStorm
3.1x baseline · 1.9k in flight
Token storm cleared
0 in flight · no session drops
WorkerOOMKilled
exit 137 · retry 2/3
Rollup completed
0 rows lost · 4m 51s
Five minutes, page to fix.
A real incident from the Pinata estate, replayed from the decision ledger: the page lands, three hypotheses are grounded against live state, one survives, and the fix is applied with rollback armed. Every line below is a ledger event.
One lab’s proof, turned into a plug-and-play install.
Today everything is wired to one estate: one stack's URLs, container names, tables and alert names. That was right for proving the approach: 9 of 11 incidents fixed end to end, 0 harmful. But a client has its own repos, monitoring, pager, chat and runtime, sometimes hundreds of each. 3am.si turns the proof into a product: a client installs it on-prem and connects to any legacy estate through a guided setup, with no integration project.
On-prem only. Nothing leaves the client's network apart from the SaaS tools they choose to connect.
Compiled binaries in signed images; an offline signed licence; our knowledge sealed in encrypted packs; a EULA.
The client's GPU, auto-detected. A CPU fallback profile so a pilot runs on any VM.
Shadow mode first (proposes, logs, touches nothing), then L1 one-tap approval, per service.
Helm chart first, plus a single-host Compose bundle for VM-only estates, built from the same images.
Source repos, monitoring and alert sources, action targets, ticketing and CMDB.
Slack, Symphony, PagerDuty, xMatters/Everbridge, Twilio voice, plus e-mail as a fallback.
Prometheus/Grafana, Splunk + ServiceNow, Datadog + Jira.
Every action, every reason and every model call logged in a tamper-evident form.
Point it at a monolith that scares you.
We’re onboarding a small number of teams running legacy core systems we can’t rewrite. The install runs on your hardware, behind your firewall, connected through a guided setup. So if you’ve got a service where the on-call rotation has learned to dread the pager, we want to see it.
Shadow mode by default: proposes and logs, touches nothing until you approve it.