Docs/Overview

Overview

A high-level look at what SQL Performance Intelligence™ does, what it avoids, and why its read-only approach is safe for production.

What This Product Is

SQL Performance Intelligence is a read-only analysis application for SQL Server. It helps you inspect CPU, waits, blocking, indexes, SQL Agent jobs, security posture, and object-level SQL details without applying automatic schema changes.

What It Does
  • Reads performance metadata, Query Store history, and object definitions.
  • Builds evidence-backed findings and exportable reports.
  • Lets you add either a local or cloud LLM for deeper interpretation.
What It Does Not Do
  • It does not create or alter your application schema automatically.
  • It does not need table-data read access to business rows for normal usage.
  • It does not start with a full setup by itself; first-run settings must be completed by the user.
Start With The Right Page

This page explains how the product works and what prerequisites matter. For the actual setup flow, go directly to the installation and quickstart guides.

Installation Guide

Use the installation guide for first-run onboarding, SQL login preparation, Query Store setup, and the initial Database plus AI / LLM configuration steps.

Quickstart Guide

Use the quickstart guide after installation to make the first SQL connection, add an AI provider, and validate the first real analysis workflow.

Before You Start
1. SQL Access

Ask your DBA for a read-only SQL login or a Windows account with the minimum diagnostic permissions. For most first-time users, the important permissions are VIEW SERVER STATE, VIEW DATABASE STATE, and VIEW DEFINITION.

2. Query Store

Ask for Query Store to be enabled on the database you want to analyze. Modules such as Query Statistics, Index Advisor, and wait-to-plan correlation work better when historical Query Store evidence exists.

3. Decide Your AI Mode

Decide early whether you want a local model such as Ollama or a cloud provider such as OpenAI, Azure OpenAI, Anthropic, or DeepSeek. That choice affects which credentials and fields you need later.

4. Keep The First Run Simple

Do not start by tuning thresholds, prompt rules, or advanced caching. First make one connection, one AI provider, and one successful analysis. Optimize later.

Why Query Store Matters

Query Store is not required for every screen, but it is strongly recommended for real analysis. Without it, the app falls back to more limited live DMV data and loses historical depth.

  • Query Statistics can rank important queries with better historical evidence.
  • Index Advisor can estimate dependency and drop risk with stronger workload context.
  • Wait Statistics can use richer query-correlation and wait-history scenarios.
  • Regression and plan-change analysis become more trustworthy than DMV-only snapshots.
Important
If Query Store is disabled, the application can still connect and some modules still work, but regression detection, plan history, and historical ranking are weaker.

If you need the DBA-facing SQL and the exact enablement sequence, continue with the installation guide.

Choose Your AI Path
Local (Ollama)
  • Best when data should stay on your own machine or inside your own network.
  • Works well for offline or controlled environments.
  • Requires a running Ollama service and an installed local model.
Cloud LLM
  • Best when you already operate OpenAI, Azure OpenAI, Anthropic, or DeepSeek.
  • Usually easier to scale and easier to standardize across many users.
  • Requires API credentials and, for some providers, endpoint or deployment fields.

You can configure multiple providers, but keep one clear default provider for daily work. For a new user, one working provider is better than many partially tested providers.

Recommended First Module Order
  1. Open Dashboard to see whether CPU, memory, IO, or TempDB pressure is already obvious.
  2. Open Query Statistics if the problem looks query-driven or workload-driven.
  3. Open Object Explorer when you need source code, object stats, dependencies, or AI Tune for one object.
  4. Open Wait Statistics or Blocking Analysis when contention is the main symptom.

When you are ready to move from concepts to execution, follow the quickstart workflow.