AI pair-DBA for SSMS 22 that proves & logs what it changes — running on your own Claude / ChatGPT / Codex account, no API key needed with Claude Code or Codex sign-in.
SQL Sage is an English-language extension you install inside SQL Server Management Studio 22 on Windows (a VSIX add-in). It brings an AI pair-DBA into the query editor: it explains errors, tunes slow T-SQL, and runs SQL safely — signed in with your own Claude (Claude Code) or ChatGPT / OpenAI Codex account, so there is no API key and no second, metered AI bill. It runs as part of SSMS on your machine; it is not a web app or an online SQL editor.
It is an independent extension built by Luma (LUMA sp. z o.o.) and is not affiliated with Microsoft, Anthropic or OpenAI.
Most AI in SSMS predicts an answer from the model. SQL Sage reads the actual server through DMVs and system catalogs, then proves and logs what it did. Four deterministic surfaces make it a real SQL Server AI assistant rather than a chat box:
- Change-Impact — before any
ALTER/DROP, see exactly what breaks: dependent views, stored procedures and triggers, inbound foreign keys, and the rows at risk. The blast radius, computed — not guessed. - Incident Mode — one-click live triage of a struggling server: the blocking chain, top waits, and the most expensive queries, with every claim cited to a real number from a DMV.
- Prove-It — proves that an AI rewrite of a query returns the same rows as the original, using a deterministic multiset fingerprint. Zero AI tokens spent — it's pure computation, not a second opinion from the model.
- Tamper-evident audit — a running, tamper-evident log of what the AI touched, with an evidence export you can hand to a reviewer or attach to a change ticket.
SQL Sage does not resell tokens and never sees your AI bill. Sign in with an account you already have:
- Claude in SSMS via the Claude Code CLI, or
- ChatGPT / OpenAI Codex in SSMS via the Codex CLI.
Pick the provider and model in the panel. With Claude Code or Codex sign-in there is no API key to paste (an Anthropic API key is optional), and no per-token charge from us. AI usage itself comes from your own Claude or ChatGPT plan.
New models appear on their own: the model list is read live from your Claude Code installation, so a new Claude model shows up without waiting for a SQL Sage release. You also choose how hard the AI thinks — a reasoning-effort selector next to the model picker offers exactly the levels your model supports (for example Low → High → Max), and /effort high raises it for a single answer. "Auto" keeps the fast default; higher effort is slower and uses more of your plan's limits.
A ScriptDom gate classifies every statement before it runs:
SELECTruns automatically so exploration stays fast.- DML / DDL stops for explicit confirmation, with the exact SQL shown in front of you before anything executes.
- Queries that reach beyond the current database — linked servers,
OPENQUERY,OPENROWSET(includingBULKfile reads),OPENDATASOURCE— always stop for confirmation, even when they look like a plainSELECT. - SQL Sage reads schema and DMVs; it never sends your query results to the model without an explicit per-session opt-in — and results you shared with one AI provider are not re-sent to another if you switch, unless you agree again.
- Run executes only in the new query window SQL Sage opened, after checking the SQL, server and database match; every step (approved → dispatched → done or failed) lands in the audit log.
- Chat history is encrypted on disk (Windows DPAPI) with a retention setting, or you can turn saving off.
- The AI can use only SQL Sage's tools: Codex runs read-only without its own shell or web search, and Claude Code's own file, shell and web tools are blocked in every SQL Sage turn. An answer stops after 12 steps or 10 minutes and tells you so.
- Need help? Copy diagnostics builds a support report you preview first — no SQL, results, server/database names or keys.
Edition-aware across box SQL Server, Azure SQL Database, and Azure SQL Managed Instance — it uses what each platform exposes and degrades gracefully where a feature doesn't exist (for example, features that depend on server-level DMVs unavailable on Azure SQL Database).
| Host | SQL Server Management Studio 22 |
| OS | Windows · x64 / Arm64 |
| Targets | SQL Server (box), Azure SQL Database, Azure SQL Managed Instance |
| AI providers | Your own Claude (Claude Code) or ChatGPT / OpenAI Codex (Codex CLI) |
- Download the signed installer from the latest release (currently 0.24.0) or from sqlsage.lumasoft.pl.
- Run it. SSMS 22 is required. When upgrading, close SSMS first.
- Open SSMS, connect to a server, and open the SQL Sage panel.
The installer is Authenticode-signed by LUMA sp. z o.o. (Certum) — since 0.21.0 end to end, including the setup stage it extracts to %TEMP% and the uninstaller, so it also runs on PCs with Smart App Control or App Control for Business. Because this is a newer publisher whose download reputation is still building, Windows SmartScreen may show "Windows protected your PC" — choose More info → Run anyway. For extra assurance, verify the SHA-256 hash published on the website and in the release notes against your downloaded file before installing.
- 30-day free trial — all features, no credit card.
- After the trial, SQL Sage keeps working as SQL Sage Free (below); AI features and the Proven & Accountable layer need a paid license. You bring your own AI account, so there is no metered AI charge from us on top.
- Current pricing (see the site for details): own it once from $39, or subscribe from $29/year.
Full, up-to-date pricing: sqlsage.lumasoft.pl
Every install starts with a 30-day free trial of everything. After it, SQL Sage keeps working as SQL Sage Free — the editor tools below, with zero AI (LLM) calls. AI features and the Proven & Accountable layer need a license.
- Free: F5 execution warnings, JOIN…ON suggestions from foreign keys, Format SQL (uses SSMS's own
formatter when available), Expand
SELECT *, Qualify object names, Execute current statement, a one-click database health check (read-only, no AI), and the proven part of Explain error — when a query fails, SQL Sage shows what the catalog proves (the missing column and the closest real name, the foreign key or CHECK behind a 547, the key columns behind a duplicate, the column size behind a truncation, the parser position of a syntax error). Values from your rows are never shown or sent. Redgate-style shortcuts (Ctrl+B, Ctrl+W · Ctrl+B, Ctrl+Q · Ctrl+K, Ctrl+Y · Shift+F5) step aside automatically when SQL Prompt is installed. - With a license: AI chat, the AI explanation and fix for a failed query, Explain / Fix / Optimize / Document, AI completions (Alt+.), and the Proven & Accountable layer — Change-Impact, Incident Mode, Prove-It.
Questions about Free: GitHub Discussions (community, no SLA).
- Website: sqlsage.lumasoft.pl
- Latest release: https://github.com/LUMASoftPL/sqlsage/releases/latest
- Discussions (Q&A): https://github.com/LUMASoftPL/sqlsage/discussions
- Issues & feature requests: https://github.com/LUMASoftPL/sqlsage/issues
- Email: support@lumasoft.pl
SQL Sage is commercial / proprietary software — © LUMA sp. z o.o. This repository hosts releases, documentation and issue tracking; the product source is not open-source. Not affiliated with Microsoft, Anthropic or OpenAI.