What is happening?
Technical evidence from the instance, not a list of alerts without context.
SQL Server 2025
Trial available
CONSULTORDBA AI · DATABASE INTELLIGENCE
Your SQL Server doesn't need another screen full of metrics. It needs to understand what's happening.
ConsultorDBA AI Health Agent combines SQL Server telemetry, Query Store, DBA expertise, historical baselines, Machine Learning and AI-assisted reasoning to detect anomalies, investigate performance problems and build evidence-based root cause hypotheses.
DEMO
Watch the AI investigate SQL Server
A SQL Server 2025 instance with multiple performance problems deliberately triggered. The agent investigates without knowing beforehand what was triggered.
REAL CASE FROM THE DEMO
From an instance with multiple symptoms to an evidence-based investigation
In the video, the agent analyzes a SQL Server 2025 instance with several performance problems deliberately triggered inside a controlled test environment. The agent has no prior knowledge of what was triggered and never queries the test environment's answer key — it investigates using only what a real DBA would have available: DMVs, metadata, Query Store, wait statistics, statistics, indexes, configuration, locks, transactions and historical telemetry. The goal isn't to show a list of alerts, but to turn hundreds of signals into evidence, root-cause hypotheses, recommendations, risk assessment and controlled remediations.
AI Investigation — demo case
✓Discover run against the instance
✓Index found with 98.7% fragmentation
✓dbo.Transactions statistics found stale
✓parallelism_wait_ratio ≈ 0.66 detected
✓runnable_tasks_per_scheduler ≈ 3 detected
✓Query Store: probable plan regression identified
Correlating root cause…
PRIMARY HYPOTHESIS
Excessive Parallelism / Scheduler Pressure, with Cost Threshold for Parallelism potentially set too low.
Confidence88%
RISK ASSESSED
Proposed actionReview MAXDOP / Cost Threshold
RiskMEDIUM
Human ApprovalYES
Blast RadiusInstance-wide
MEASURED OUTCOME
INCONCLUSIVEDurationno measurable change
CPUno measurable change
PAGEIOLATCHno measurable change
ResultNO_MEASURABLE_CHANGE
What's the most likely cause?
With the evidence that supports it — and the evidence that contradicts it — made explicit.
Did the fix actually improve SQL Server?
Measured before and after, never assumed just because the SQL command ran without error.
During this test, the agent never queries the table that holds the test environment's correct answer — it investigates the way an outside DBA would. That's what makes its results measurable.
True PositivesFalse PositivesFalse NegativesRoot Cause AccuracyRemediation Effectiveness
READ-ONLY DISCOVERY
FREE · LIMITED SLOTS
SQLDISCOVERY AGENT · FREE SQL SERVER HEALTH CHECK
See what deserves attention before it becomes a performance problem.
SQLDiscovery Agent performs a read-only Health Check against one SQL Server instance and database, reviewing configuration, waits, Query Store, execution workload, indexes, statistics, files, transaction log, backups, TempDB and blocking, then generates a severity-ranked HTML report. No changes to your instance.
READ-ONLYSQL SERVER 2012–2025HTML REPORT30-DAY TRIAL
WHAT IT DOES
From connection to report, in minutes
- 1
Connect
Select the instance and database.
- 2
Discover
Reviews configuration and performance signals without changing SQL Server.
- 3
Prioritize
Generates an HTML report with Critical, High, Medium and Low findings.
WHAT IT CHECKS
Six technical areas, one read-only pass
The same judgment a senior DBA would apply in a manual review — just in minutes, and consistently.
Instance & Database Configuration
Cost Threshold for Parallelism, MAXDOP, Max Server Memory, Optimize for Ad Hoc Workloads, PAGE_VERIFY, AUTO_SHRINK, AUTO_CLOSE, Auto Statistics, TRUSTWORTHY, DELAYED_DURABILITY, database-scoped MAXDOP and Legacy Cardinality Estimator.
Workload & Waits
CXPACKET, PAGEIOLATCH, WRITELOG, LCK_M, RESOURCE_SEMAPHORE and any blocking present at read time. Waits are cumulative since the last SQL Server restart, not just from the Health Check itself.
Query Store & Plan Cache
Query Store OFF or READ_ONLY, ALL capture mode, longest-running SQL, plans for the selected database, and single-use ad hoc plans.
Indexes & Statistics
Missing indexes from the DMVs with their seeks and estimated impact, indexes with no reads and heavy writes, fragmentation on large indexes, statistics with a high number of modifications, and low Fill Factor as information only — never as an automatic recommendation.
Files, Log & TempDB
Data and log file growth, VLF count, why the log can’t be reused, and TempDB file count and size differences.
Backup & Recoverability
Last Full, Differential and Log backups, recovery model, and the risk when a database is in FULL recovery without an adequate Log Backup strategy.
An index should not be dropped just because it shows no reads in this point-in-time snapshot.
DEMO
Watch a complete SQL Server Health Check in under 4 minutes
From connecting to SQL Server to the severity-ranked HTML report.
Your instance doesn’t change
SQLDiscovery Agent analyzes your instance and generates evidence. It doesn’t change configuration or modify data.
During the Health Check, SQLDiscovery Agent never runs:
ALTERUPDATEINSERTDELETECREATEDROP
It doesn’t change server or database configuration, indexes, statistics, passwords, or business data.
If SQL Server forces a password change for the user account, SQLDiscovery Agent stops and asks you to change it from SQL Server Management Studio before continuing.
It only needs this
- SQL Server version
- Instance
- Database
- SQL user and password
- Folder to save the report
THE NEXT LEVEL
The Health Check is the start. Understanding root cause is the next level.
SQLDiscovery Agent takes a technical snapshot of your instance and shows you what deserves attention. If you need to go beyond the snapshot — learn the instance’s historical behavior, detect anomalies, correlate evidence, investigate root cause and validate remediations — meet ConsultorDBA AI Health Agent.
TWO TOOLS, ONE ECOSYSTEM
SQLDiscovery Agent and AI Health Agent
SQLDiscovery Agent
- Point-in-time Health Check
- Read-only
- 1 database
- HTML report
- Configuration + performance discovery
- No remediation
- No historical baseline
AI Health Agent
- Continuous intelligence
- Historical telemetry
- Baseline learning
- Machine Learning
- Root Cause Analysis
- AI Investigation
- Controlled remediation
- Before/after validation
MORE THAN DETECTING PROBLEMS
More than detecting problems
SQL Server can generate thousands of metrics. ConsultorDBA AI Health Agent turns those signals into evidence, anomalies, root cause and concrete actions — it isn't a DMV followed by a rule followed by an alert.
OBSERVE
LEARN
DETECT
CORRELATE
INVESTIGATE
DIAGNOSE
RECOMMEND
ASSESS RISK
REMEDIATE
VALIDATE
LEARN FROM OUTCOME
ARCHITECTURE
A hybrid architecture, not a rules engine in disguise
DBA rules, historical telemetry, baseline learning, feature engineering, Machine Learning, anomaly detection, Query Store intelligence, root cause correlation, AI reasoning, knowledge retrieval, remediation risk analysis, controlled remediation and outcome learning — all working together, not in isolation.
SQL SERVER
NON-INVASIVE COLLECTION
TELEMETRY MEMORY
FEATURE ENGINEERING
DBA RULES
MACHINE LEARNING
QUERY STORE
ANOMALY DETECTION
QUERY REGRESSION
ROOT CAUSE CORRELATION
AI INVESTIGATION
KNOWLEDGE RETRIEVAL
REMEDIATION RISK ENGINE
HUMAN APPROVAL
CONTROLLED REMEDIATION
BEFORE / AFTER TEST
OUTCOME LEARNING
1 · NON-INVASIVE COLLECTION
Evidence without putting your instance at risk
The agent gathers evidence through DMVs, Query Store, waits, metadata, configuration, indexes, statistics, locks, blocking, transactions, TempDB, I/O, performance counters and Extended Events when appropriate. It is designed to minimize impact on the instance and avoids reading business data when not necessary.
DMVsQuery StoreWaitsMetadataConfigurationIndexesStatisticsLocksBlockingTransactionsTempDBI/OPerformance countersExtended Events (when applicable)
2 · BASELINE LEARNING
Learns how this specific instance behaves
The agent builds percentiles (P50, P75, P90, P95, P99), historical averages and hourly and day-of-week profiles for each instance. 68% CPU may not be critical in absolute terms, but it can be completely abnormal for that instance at that time of day.
CPU — CURRENT INSTANCE
CPU CURRENT68%
Historical P5034%
Historical P9546%
Deviation+48%
Illustrative example. Real percentiles depend on each instance's accumulated historical telemetry.
3 · QUERY STORE INTELLIGENCE
Identifies real degradations, not just different plans
The agent analyzes Query Store to identify real performance degradations. A different plan does not mean a bad plan — there has to be measurable impact for it to qualify as a regression.
BEFORE
Plan18
Duration182 ms
Logical Reads4,282
AFTER
REGRESSION SCORE: 98%Plan23
Duration3,918 ms
Logical Reads824,812
QUERY 0x91AB... — Illustrative example, not real customer telemetry.
4 · TELEMETRY MEMORY
Not a single snapshot of the server
Every analysis keeps local historical telemetry, so current behavior can be compared against that same instance's historical behavior — not against a generic threshold applied to any server.
CURRENT
HISTORICAL BEHAVIOR
5 · FEATURE ENGINEERING
From raw values to useful technical signals
Wait times, I/O, CPU, execution plans, blocking and statistics are transformed into ratios, deviations, changes, trends and workload indicators — the signals that Rules and Machine Learning can actually use, not raw numbers.
RAW SIGNALS
Wait timesI/OCPUExecution plansBlockingStatistics
ENGINEERED FEATURES
RatiosDeviationsChangesTrendsWorkload indicators
6 · DBA RULES ENGINE
Deterministic DBA knowledge, not replaced by AI
The agent contains deterministic administration knowledge: misconfiguration, statistics, indexes, autogrowth, Query Store, database options, parallelism and TempDB. AI does not replace these rules — it complements them.
MisconfigurationStatisticsIndexesAutogrowthQuery StoreDatabase optionsParallelismTempDB
7 · MACHINE LEARNING / ANOMALY DETECTION
Detects changes a fixed threshold can miss
The system uses statistical analysis and Machine Learning to identify spikes, change points, persistent deviations, workload shifts, query latency anomalies, I/O anomalies, wait anomalies and blocking changes. Machine Learning does not replace DBA rules — it is a hybrid architecture.
SpikesChange pointsPersistent deviationsWorkload shiftsQuery latency anomaliesI/O anomaliesWaits anomaliesBlocking changes
RULES
BASELINE
MACHINE LEARNING
8 · ROOT CAUSE CORRELATION
The visible symptom isn't always the cause you need to fix
The agent correlates waits, queries, plans, statistics, indexes, I/O, concurrency, configuration and historical behavior to build the real cause-and-effect chain, instead of stopping at the most visible symptom.
STALE STATISTICS
CARDINALITY ESTIMATION ERROR
POOR EXECUTION PLAN
EXCESSIVE READS
I/O PRESSURE
PAGEIOLATCH
SLOW QUERY
WaitsQueriesPlansStatisticsIndexesI/OConcurrencyConfigurationHistorical behavior
9 · AI INVESTIGATION
Actions, evidence and hypothesis — not the model's internal reasoning
The investigation is shown the way a DBA would review it: what was checked, what evidence was found, and which hypothesis best explains the symptoms — never the model's internal chain-of-thought.
AI Investigation
✓Workload baseline loaded
✓Current anomalies identified
✓File latency analyzed
✓Top I/O queries found
✓Query Store history compared
✓Execution plans reviewed
✓Statistics state inspected
Investigating I/O pressure…
PRIMARY HYPOTHESIS
Execution plan regression associated with stale statistics.
Confidence93%
10 · KNOWLEDGE RETRIEVAL
Instance evidence, with more technical context
The agent can complement the evidence collected from SQL Server with a specialized technical knowledge base — instance evidence plus DBA knowledge gives better context for the hypothesis, without querying confidential external information.
11 · CONFIDENCE SCORE
Every hypothesis shows its own evidence
A root cause hypothesis always includes supporting evidence, contradicting evidence (if any), and missing evidence — not just an unsupported percentage.
ROOT CAUSE
Plan Regression · 94%- Plan changed
- Duration +1,800%
- Reads +4,200%
- Statistics heavily modified
- None detected
12 · DIAGNOSTIC AREAS
Full technical coverage, not a single angle
The agent covers the areas where most SQL Server production performance problems actually originate.
Queries
- Expensive queries
- Slow queries
- CPU
- Logical reads
- Physical reads
- Query Store
- Execution plans
- Regressions
Indexes
- Missing
- Duplicate
- Overlapping
- Unused
- Fragmentation
- Page density
- Key lookups
- Excessive index maintenance
Statistics
- Stale statistics
- Modification counters
- Cardinality problems
- Estimated vs. actual rows
Blocking & Deadlocks
- Blocking sessions
- Head blockers
- Blocking chains
- Long transactions
- Deadlock evidence
TempDB
- Configuration
- File sizes
- Growth
- Usage
- Version store
- Spills
- Contention signals
I/O
- File latency
- Reads
- Writes
- Workload correlation
- PAGEIOLATCH
- WRITELOG
Parallelism
- MAXDOP
- Cost Threshold
- CXPACKET
- CXCONSUMER
- Worker/scheduler pressure
Memory
- Memory grants
- Waits
- Spills
- Query memory pressure
Triggers
- Trigger presence
- Potential write amplification
- Blocking correlation
- DML impact
Database Configuration
- Filegrowth
- Query Store
- Database options
- Performance-relevant configuration
13 · REMEDIATION RISK ENGINE
Detecting a problem and executing a fix are two different decisions
Every proposed remediation includes risk level, expected benefit, confidence, blast radius, reversibility, whether human approval is required, a validation plan and a rollback plan. The agent does not promise universal results.
PROPOSED ACTION
UPDATE STATISTICSConfidence94%
Expected BenefitHIGH
RiskLOW
Requires ApprovalYES
14 · CONTROLLED REMEDIATION
The LLM does not execute arbitrary SQL directly
The agent does not hand absolute control to the AI model. Every remediation goes through a risk engine and requires human approval before it runs — the user stays in control at every step.
AI Recommendation
Risk Engine
Approved Action
RemediationService
15 · BEFORE / AFTER VALIDATION
We don't assume a fix worked. We measure it.
BEFORE
Duration3,921 ms
Logical Reads912,821
AFTER
Duration188 ms
Logical Reads8,123
Illustrative example. Results depend on the workload and root cause.
16 · OUTCOME LEARNING
A history of which actions actually worked for this instance
The agent records the full trail — diagnosis, recommendation, action, before, after and outcome — to build a history of which remediations were effective on that specific instance. This is Outcome Learning (or Remediation Outcome Memory), not automatic retraining in production and not an agent that modifies its own code.
Diagnosis
Recommendation
Action
Before
After
Outcome
17 · SECURITY
Intelligence without handing unrestricted control to AI
- Non-invasive diagnostics, designed to minimize impact on the instance
- Controlled timeouts on every observation query
- Observation queries only — never modifies data during diagnosis
- Does not store passwords
- Does not store connection strings insecurely
- Avoids accessing business data rows
- SQL text can be configured per client policy
- The LLM does not execute arbitrary SQL on its own
- Controlled remediation, never automatic without approval
- Human-in-the-loop on every remediation action