"An increase at this time of day is normal." "Orders dip at month-end." Business context can't go into a threshold, but it can go into memory. Facts recorded by AI and members inform every decision from then on.
A spike in connections during the 3 a.m. daily backup is normal
Recorded by a user
pending in payments.status means waiting on the payment processor, and normally clears within a few minutes
Recorded by AI
Order volume tends to dip at month-end
Recorded by a user
The rise in connections around 3 a.m. matches the daily backup in memory. Nothing is wrong.
The same false alarm gets sent again and again
The people who know the business aren't the ones who set up monitoring
You loosen thresholds after every false alarm, until real anomalies slip through
AI saves facts learned in investigations to memory. It rewrites entries that have gone stale and deletes ones that were wrong.
People on the business side know the business context best. They can add a one-sentence entry from the console, and it's used just like the ones AI records.
Memory is referenced in scheduled judgments, cause investigations, patrols, and chats.
Each memory shows whether it was recorded by AI or by a member, and when.
No special format needed. Write it in one or two sentences, the way you'd tell a colleague.
Changes that happen at set times, like "a spike in connections during the 3 a.m. daily backup is normal."
Changes that previously turned out to be false alarms, and why.
What a column's values represent for the business, and which states are normal.
Context known to people on the business side, like "orders dip at month-end."
All you need is a database connection string or an account connection. Nothing to install on your servers.
AI records
AI saves facts learned in investigations and chats to memory.
Members add to it
Add business context or notes on AI's entries from the console.
It informs later decisions
From the next judgment and investigation on, decisions take memory into account.
Scheduled jobs
Learn that nightly batch load is normal
Record once that values spike during the nightly batch, and later judgments take it into account. No more getting the same alert every night.
Business cycles
Don't flag the month-end dip in orders
Once someone on the business side adds "orders dip at month-end," AI takes that context into account when judging.
Anomaly detection
AI judges whether something is abnormal based on value trends, past alerts, and your project's memory.
Automatic investigation
When something looks abnormal, AI digs into the real data to trace the cause, and records its conclusion and evidence.
Patrols
Even on days with no anomalies, AI reviews monitors, database descriptions, and memory, and maintains only what needs it.
Build operational knowledge
into your monitoring.
What you record takes effect from the very next judgment.