Microsoft’s Oracle-to-Postgres code converter now works in your own repo
A public preview from Microsoft lets GitHub Copilot assess and rewrite Oracle-dependent application code for Azure PostgreSQL, directly in the existing Git repository, with reports for review.

Microsoft has put AI-assisted Oracle-to-PostgreSQL application code migration into public preview, according to an Azure Updates entry dated November 17. Per Microsoft’s tutorial, GitHub Copilot assesses and rewrites Oracle-dependent code in place, inside the developer’s own Git repository.
The Azure Updates page itself returned no readable text when we fetched it, so the details below come from Microsoft Learn’s tutorial for the feature, which was updated on the same day, and from earlier Microsoft posts. Microsoft’s own wording is attributed throughout.

Two commands, two reports, no code copying
The workflow runs through a skill called migrate-oracle-application, part of a Postgres Skills plugin used with GitHub Copilot in Visual Studio Code. The tutorial says the same flow works in GitHub Copilot CLI, which uses its own saved PostgreSQL connection profile.
A developer first runs the skill with mode=assess. It looks for Oracle-specific SQL, drivers, connection settings, ORM mappings and routine calls, checks the destination PostgreSQL objects, and evaluates the existing build and tests. The result is a report at .github\migrate-oracle-app\assessment\assessment-report.md. This step is read-only for the database and does not change source files by default.

A second run with mode=convert asks for approval to edit source files, converts the code in place and uses build and test results to validate it. It writes conversion-report.md and file-diffs.md, and leaves the changes uncommitted in the working tree. The skill does not stage, commit or push anything. A combined mode=assess-and-convert skips the review pause.
What you need, and what it will not touch
The prerequisites are fairly heavy. The tutorial lists Copilot access with a supported code-capable model, the Microsoft PostgreSQL extension, a Git repository with a clean working tree, the application’s own build tools, Node.js 22 or later, Docker for integration tests on synthetic data, and a saved connection to a nonproduction PostgreSQL database.

Test generation is opt-in through options=tests. Creating persistent database objects and running write tests against the target are separate opt-ins that are off by default. The tutorial also states that the tool does not migrate production data, which has to be planned separately. The older “Application Migration” experience in the VS Code extension remains supported, but Microsoft calls the skill-based route the recommended one.
The preview documentation does not give pricing, regional availability or a list of supported languages and frameworks in the material we could read. Earlier Microsoft material for the broader migration tooling described it as free, but this specific skill’s costs, including the Copilot subscription it relies on, are not spelled out.
A shift from copy-and-convert to evidence-driven agents
The change matters most in how the work is checked. In the first version of the tooling, described in a March 2026 Microsoft PostgreSQL blog post, developers copied source code into an application_code folder inside a migration project and ran “Migrate Application.” The same post described a multi-agent schema converter in which a critic agent ran the converted DDL against a scratch PostgreSQL database and retried on failure.
The new flow applies similar thinking to application code. It builds a baseline from the repo’s existing tests, runs conversions against disposable PostgreSQL containers in Docker, and reports unavailable validation as unresolved rather than passed. Microsoft’s tutorial is blunt about the limits: AI-assisted conversion “can produce incorrect or incomplete changes,” and a successful build or test run does not prove every Oracle dependency is resolved.
That caution fits the hard part of this problem. Oracle behavior around null handling, transactions, generated identifiers and stored routines differs from PostgreSQL in ways that compile fine and fail later. The tutorial flags exactly these areas for human review, along with generated tests and their assertions.
The tool also sits in a wider push by Microsoft to make PostgreSQL the landing place for Oracle workloads. A day earlier, our coverage of Azure HorizonDB’s Postgres knowledge-graph feature showed the same database family getting AI tooling on the data side. Microsoft’s March post named lower license costs, less vendor lock-in and extensions such as pgvector as reasons to move.
None of Microsoft’s claims about speed or accuracy are quantified in the sources we read, and none have been independently verified. Teams weighing the preview should treat the generated reports as a starting checklist, and plan for review and performance testing before anything reaches production.
Sign Up for the Daily AI Pulse
One email a day. All the stories that matter.
By signing up, you agree to our Terms of Use and acknowledge the data practices in our Privacy Policy. You may unsubscribe at any time.
Leo MartinsAI Tools, Prompts & Practical AI ExpertLeo Martins is the AI Tools & Practical AI Expert at Aiholics, focused on helping readers use artificial intelligence to improve productivity, creativity and everyday work.
You may also like

Copilot code review can now be switched on or off by API, from enterprise down to repo

IBM and Microsoft bundle 6,000 pre-built AI agents into Imagine Enterprise

GPT-6.1 Sol lands in Microsoft Foundry, as Routines and Tool Search go GA
