ARIKI — Datascout AI
The night your pager didn’t ring.
AI copilots that watch, explain and write — on local models, next to your data. Private by default.
Follow one night
02:47
Anomaly Watch notices.
Somewhere in a data centre, a delivery-message curve starts to drift. No threshold has tripped. No human is awake. Anomaly Watch — classic ML on the measurement curves you already collect — marks the drift and opens a case.
02:48
The alert explains itself.
One minute later there is already an answer to the question every on-call engineer asks first. Not a runbook link. Not a wall of labels. Three grounded sentences.
Anomaly Watch · backlog rising — pipeline wsp-delivery-east
The backlog on the delivery queue began climbing at 02:31, about four times faster than the usual nightly batch. Two related signals moved with it: consumer lag on the same pipeline, and a retry spike from one downstream endpoint. Start with the consumer log on node 3 — the pattern matches a stuck worker, not a traffic surge.
grounded in: your alert · your metrics · your logs — answered by a model that never leaves your network03:12
Nobody’s phone lit up.
The breach never happened.
06:00
Stack Doctor does its rounds.
While the kettle boils, Stack Doctor walks the whole stack — disks, certificates, queues, snapshots — and writes down the one thing actually worth a human’s attention.
Stack Doctor — morning rounds
06:00- Disk headroomall nodes above 25% free
- Certificate expirynearest expires in 71 days
- Queue depthdrained back to baseline by 03:40
- Snapshots & backupslast run verified, restorable
- ILM policywarn: hot tier will overflow in ~9 days — suggested fix attached
08:30
The report reads itself out.
Morning light. The monthly service report doesn’t wait for someone to write an executive paragraph — it arrives with one, grounded in the month’s own numbers.
Service Report — May 2026
✦ Management Summary — written by ARIKI
Availability held at 99.97% across all monitored services. One anomaly was flagged and resolved before customer impact; time-to-explanation stayed under two minutes. Storage growth suggests a hot-tier adjustment this quarter — details and the suggested change are in section 4.
09:00
The analyst just asks.
No query language, no dashboard safari. Plain language in, a real query out — visible, auditable, yours to edit.
{ "query": { "bool": { "filter": [
{ "term": { "delivery.status": "FAILED" } },
{ "range": { "@timestamp": { "gte": "now-1d/d", "lt": "now/d" } } }
] } } }147 hits · 0.21 s · query shown, nothing hidden.
09:01 — full daylight
Every model in this story ran inside their walls.
The watching, the explaining, the writing — all of it on local models, next to the data. Nothing in this night ever left the building. That is not the industry default.
ARIKI
Runs the models next to your data — your hardware or ours, your tenant, your network. Nothing leaves.