Oracle 23aiOne database. Logs, vectors and anomaly detection.

Find the failing service before your users do.

Obsrv pulls logs from every region, flags error spikes against a rolling baseline, and finds related incidents by meaning. No search cluster to run.

obsrv / error-activity · 24h
Live

Events · 24h

—

across 4 regions

Errors · 24h

—

waiting for API

Anomalies

—

vs 24h baseline

Ingest · 5 min

—

events received

Why one database

Replace a five-tool stack with one.

Typical stack

  • Elasticsearch cluster
  • Kibana dashboards
  • Separate ML anomaly service
  • Vector store for similarity
  • Sync jobs to keep them aligned

Obsrv on Oracle 23ai

  • JSONlogs stored as native JSON, queried in SQL
  • VECTOR(768)embeddings beside the row they describe
  • Materialized viewhourly error counts, refreshed on commit
  • Window functionsmean + 3σ baseline per service

How it works

Four steps, all inside SQL.

  1. 01

    Ingest

    Services POST JSON logs to a single endpoint.

  2. 02

    Embed

    Every message becomes a 768-dimension vector.

  3. 03

    Detect

    Window functions compare the last hour to a 24-hour baseline.

  4. 04

    Search

    VECTOR_DISTANCE ranks related errors by meaning.

POST /api/logs/ingest
{
  "timestamp": "2026-11-08T10:42:07+05:30",
  "service":   "payment-gateway",
  "region":    "India",
  "severity":  "E",
  "message":   "payment gateway timeout",
  "meta": { "request_id": "req-482910" }
}
$ tail -f obsrv/events (UTC)Live
  • waiting for events…

Regions

Four regions, each in its own time.

An event at 02:00 in Tokyo isn’t 02:00 in San Francisco. Logs are stored in UTC and shown in the timezone of the region that produced them.

See what’s breaking right now.