Key takeaway
- Compare Microsoft Copilot for databases in Azure SQL, SSMS, VS Code and Fabric: what each does, permissions, licensing, limits and an adoption sequence.
Copilot for databases now means several different Microsoft experiences, not one tool: Microsoft Copilot in Azure for Azure SQL Database, GitHub Copilot in SQL Server Management Studio (SSMS), GitHub Copilot for the MSSQL extension in Visual Studio Code, and Copilot in the SQL database workload of Microsoft Fabric. Each has its own prerequisites, licensing and permission model, so the first decision is which surface fits your estate.
For enterprise teams the useful question is control, not novelty: what the assistant can see, what it can run, and who approves changes. This guide sets out what each surface does according to Microsoft Learn, a decision table, governance points and an adoption sequence. For the wider delivery picture, see our guide to Azure Copilot and GitHub Copilot for DevOps teams, and for the Dynamics side of the Microsoft estate, see Microsoft Dynamics Copilot.
What does Copilot for databases mean in 2026?
The old idea of a single code-completion tool does not match Microsoft's current documentation. Microsoft Learn describes four separate experiences, summarised below. Names, previews and availability change often, so confirm each on the linked Microsoft page before you commit.
| Surface | What it is, according to its site | Where you use it | Prerequisite to confirm |
|---|---|---|---|
| Microsoft Copilot in Azure with Azure SQL Database | A capability of the Azure Copilot experience that gives a conversational interface for managing and troubleshooting Azure SQL Database; overview | Azure portal | Copilot enabled in your Azure tenant |
| GitHub Copilot in SSMS | Chat, inline chat, completions, next edit suggestions, database instructions and, from SSMS 22.7, Agent mode; overview | SQL Server Management Studio | GitHub account with Copilot access; AI Assistance workload |
| GitHub Copilot for the MSSQL extension | Schema-aware chat with @mssql, agent mode, plan mode, slash commands and inline completions; overview | Visual Studio Code | GitHub Copilot subscription and the MSSQL extension |
| Copilot in Fabric SQL database | Chat pane, code completion and Fix and Explain quick actions; overview | Fabric portal query editor, SSMS, Visual Studio Code | Tenant switch enabled; paid F2 or P1 or higher capacity |
What can Copilot do for database administrators and developers?
Microsoft Learn documents these tasks, grouped by role. None of them replaces review by someone who understands the database.
Administrators: diagnose and explain
In Azure SQL Database, Copilot answers questions using database context, dynamic management views and catalog views. Sample skills in Microsoft's prompt list include active user connections, blocking session analysis, anti-pattern query analysis, automatic tuning analysis and connection string generation. Microsoft's own example prompts include which service tier to use and why a database was unavailable.
Developers: write, explain and fix T-SQL
SSMS Copilot answers questions about SQL Server, Azure SQL Database, Azure SQL Managed Instance and SQL database in Fabric, and helps write, document, explain and fix T-SQL from natural language. The MSSQL extension adds schema suggestions, mock data and seed scripts, ORM migration scripts, plain-language explanations of stored procedures and checks for issues such as SQL injection risks or excessive permissions. Fabric adds Fix and Explain quick actions in the query editor.
Multi-step work: agent modes
SSMS Agent mode lets you state a goal and have Copilot execute queries, read execution plans and propose schema changes, with approval before each query or command. Microsoft's scenarios include creating and testing a stored procedure, analysing a slow query, reviewing the SQL error log and finding permission gaps. Microsoft advises making such changes in a test or development environment first.
How does Copilot treat permissions and data?
This is the section security and data leaders should read first. The documented behaviour differs by surface, which is why a single policy statement is not enough.
| Question | Azure SQL Database (Azure portal) | SSMS (Visual Studio Code agent mode uses the extension's connection and credentials) | Fabric SQL database |
|---|---|---|---|
| What access does it have? | Only resources you can access, and only actions you have permission to perform | Queries run under your login; a query on a table you cannot read cannot run | Schema, your prior messages, queries you ran and their errors; not table data |
| Can it change data? | Microsoft describes it as answers and self-help; review output before production use | Agent mode asks approval for each query or command; the security boundary is SQL Server permissions | Read-only mode suggests code; read and write mode needs your approval |
| Training and retention | Prompts and responses are not used to train the foundation models, per Microsoft's FAQ | Microsoft Learn states GitHub Copilot in SSMS does not retain prompts or responses or use your data to train models | Prompts and responses are not used to train foundation models, per Microsoft |
Sources: Azure SQL Copilot FAQ, SSMS Agent mode and Fabric SQL database Copilot. The practical rule is to give the human or service identity the least privilege it needs, because Copilot inherits that identity's reach. For wider controls on autonomous actions, see agentic AI risk management.
Which Copilot should you choose?
| Your situation | Best fit | Why | Check before you commit |
|---|---|---|---|
| Estate on Azure SQL Database, portal-first operations team | Copilot in Azure with Azure SQL Database | Answers in the portal using database diagnostics | Tenant enablement; confirm preview or general availability status |
| DBAs who live in SSMS on SQL Server or Azure SQL | GitHub Copilot in SSMS | Chat, completions and Agent mode in the tool they use | GitHub account and plan; SSMS version |
| Application developers writing T-SQL and migrations | GitHub Copilot for the MSSQL extension | Schema-aware chat and agent mode beside application code | Subscription; note that inline completions do not see your connected schema |
| Operational data in Fabric SQL database | Copilot in Fabric SQL database | Built into the query editor with Fix and Explain | Paid capacity, region and tenant settings; trial SKUs are not supported |
Teams on Azure Data Studio should note that Microsoft Learn states it retired on 28 February 2026, with Visual Studio Code and the MSSQL extension as the recommended replacement (Microsoft notice). Teams planning governed data and AI platforms can also see how ElixirData Context OS keeps governed context, versioned policy and decision traces in one decision layer above AI agents.

What do Copilot for databases cost and what limits apply?
Licensing differs by surface, and Microsoft revises it, so treat the following as the position when we read the pages and confirm it with your Microsoft account team. The Azure SQL Database Copilot FAQ states that its skills are offered at no additional cost. SSMS and Visual Studio Code use a GitHub account with Copilot access, and Microsoft documents a free plan for SSMS. Fabric requires a paid capacity of F2 or P1 or higher and a tenant administrator switch, and outside the US or France it is disabled by default unless the administrator allows data to be processed outside the capacity's region.
Known limits from the documentation: Copilot in Fabric SQL database cannot edit an existing query in the editor through the chat pane, may be inaccurate when asked to evaluate data because it sees only schema, and its chat pane is not available when Private Link is enabled and public access is disabled in the tenant setting. Microsoft states that AI output can contain mistakes and must be reviewed.
How should an enterprise adopt Copilot for databases?
A staged sequence keeps risk visible. Each step should have an owner and an exit criterion.
- Inventory. List engines, tools and the Microsoft surface each team already uses.
- Scope identities. Create read-only logins for exploration and keep write rights separate, because Copilot acts as the signed-in user.
- Pilot in non-production. Use development and test databases, with a short list of approved prompts and review steps.
- Add review gates. Treat generated T-SQL like any pull request: peer review, tests and a recorded approval.
- Record decisions. Keep an audit trail of which statements were suggested, approved and run, using AgentGRC, which keeps autonomous agent actions inside policy, with the audit trail risk teams need.
- Widen deliberately. Move to production-adjacent use only when access reviews and tests show the controls hold.
Related delivery practices are covered in Database DevOps, Data DevOps and why data quality matters. For building and governing agents across such workflows, see ElixirClaw, an agentic OS platform to build, orchestrate and govern AI agents and automation workflows while keeping humans in control.
When is Copilot not the right tool?
| Situation | Why caution is needed | Safer approach |
|---|---|---|
| Unreviewed changes to production schema | Microsoft states the security boundary is SQL Server permissions and that output can be wrong | Generate in test, review, deploy through your pipeline |
| Questions that need the actual data values | Fabric Copilot sees schema only and may be inaccurate when evaluating data | Run reviewed queries and validate the results yourself |
| Highly regulated data movement | Fabric processing outside region needs an administrator decision | Keep the tenant setting off until legal review is done |
| Teams without a data owner | Nobody can judge whether suggestions are correct | Assign an accountable owner first |
To map one database decision your agents should make, and the context and policy it depends on, book a Context Workshop. See what happens in the workshop.
Frequently asked questions
Does Copilot in Azure SQL Database see my table data?
According to Microsoft Learn, Copilot in Azure SQL Database answers from sources such as public documentation, dynamic management views, catalog views and Azure supportability diagnostics, and it only has access to resources you have access to. Copilot for SQL database in Microsoft Fabric uses object names and key metadata, not the data in tables.
Can Copilot change my database on its own?
Not without approval, according to Microsoft Learn. In SSMS Agent mode, Copilot asks for approval before each query or command, and Microsoft states that the security boundary is SQL Server permission enforcement, not Copilot's approval prompt. In Fabric, read-only mode suggests code for you to run, and read and write mode needs your approval.
What do I need to use Copilot in SQL Server Management Studio?
According to Microsoft Learn, SSMS 22.2 or later with the AI Assistance workload is needed for completions, SSMS 22.7 or later for Agent mode, and a GitHub account with Copilot access (a free plan is documented). Confirm current requirements on the Microsoft page before rollout.
Is Azure Data Studio still an option for Copilot?
No. Microsoft Learn states that Azure Data Studio retired on 28 February 2026 and recommends Visual Studio Code with the MSSQL extension, which includes GitHub Copilot integration.
Should production databases be Copilot-enabled for everyone?
Start in development and test, grant least-privilege logins, review every generated statement, and widen access only after the controls are proven. Microsoft advises testing changes in a non-production environment first.
Follow XenonStack on LinkedIn: XenonStack on LinkedIn for new guides on agentic AI, data and platform engineering.
Explore the XenonStack platform family: ElixirData · ElixirClaw · ElixirHub · NexaStack · MetaSecure · Neural AI · AgentGRC · AgentSRE · AgentAnalyst · AgentInstruct · AgentRAI · AgentScrum · AgentSearch · AgentForce · AI Aviator · SAIF Aviator · CohortIQ · XenonStack AI · XenonStack UAE