Ptah

Ptah compares the state you want with the state a database has, then produces a change you can inspect before it runs. Schema work can land as versioned SQL files or as a direct apply. Persistent model output uses a separate candidate, verification, cutover, and rollback lifecycle.

The default quick start creates and verifies a SQLite database in about five minutes. It needs no database server, Docker, or Go toolchain.

Schema changes and persistent inference state use separate review and cutover paths.

What to notice: Schema changes converge a desired schema and a database through versioned files or direct apply. Inference changes keep the active generation available until a candidate passes verification and cutover.

Open full sizeDownloadView source

See what Ptah produces

These are generated from committed fixtures by the same commands the linked pages document—not mockups or conceptual diagrams.

Entity-relationship diagramRender the desired schema as Mermaid, DOT, or SVG.

Browsable schema referenceExport a self-contained HTML document with counts, navigation, a diagram, and table detail.

Live drift viewServe the declared schema beside a live drift verdict that updates as the database changes.

Migration safety reportKeep a reviewable HTML safety verdict beside a generated migration pair.

Start with your situation

Get a first resultApply one reviewed schema plan to a disposable SQLite database and read the live result back.

Choose how schema changes landDecide between versioned migrations and direct schema changes before adopting either workflow.

Change persistent inference stateBuild and verify a candidate generation before switching the active embedding or model output.

Three product domains

Versioned migrations

Keep ordered SQL files in your repository when a change must be reviewed, replayed across environments, and represented in deployment history.

Start with a versioned migration or open the versioned workflow overview.

Direct schema changes

Keep a desired schema and let Ptah plan the SQL that makes a target database match it. This path suits a database with one clear owner and changes that do not need a committed migration file.

Run the default quick start or open the direct workflow overview.

Inference migrations

Keep the active generation available while Ptah builds, catches up, and verifies a candidate generation. Cutover changes the active pointer only after the candidate passes the required checks.

Check whether inference migrations apply.

Common tasks

Adopt an existing databaseInspect what exists, then baseline migrations or adopt a desired schema without recreating objects.

Generate a migrationTurn the difference between a reviewed desired schema and a live database into migration files.

Detect schema driftCompare a live database with the desired schema and use the exit code as a CI gate.

Check database supportRead the generated release-line census, support policy, and engine-specific guidance.

Embed Ptah in GoUse the stable public packages for parsing, planning, rendering, and migration work.

Connect an AI clientExpose a constrained Ptah tool surface over MCP and choose what the client may change.

Evidence and compatibility

Support levels describe what this repository exercises, not what a server is allowed to do. The database support matrix is generated from the same release-line declaration used by CI.

Atlas compatibility is a separate journey through the ptah-compat binary. The Atlas overview distinguishes default compatibility, strict Community Edition mode, native Ptah, measured parity, and retained divergences. The native ptah binary does not register Atlas command paths.

Ptah is an independent MIT-licensed implementation. The license boundary records the provenance policy.