@sedrino/db-schema

Migration-first schema planning for AI-authored SQLite/libSQL databases.

This package is aimed at a workflow where:

1. AI or humans author migrations in a deterministic TypeScript DSL
2. the planner materializes the next schema snapshot JSON
3. the compiler emits generated artifacts such as Drizzle schema code

Current scope

The first version focuses on the planning layer:

• versioned schema document types
• migration DSL with deterministic operation recording
• schema materialization from migration history
• rebuild-aware SQLite migration emission for supported table-shape changes
• Drizzle source generation for a narrow SQLite + Temporal-aware subset
• inferred Drizzle soft-relations generation from foreign keys
• a libSQL-compatible apply runner with migration/state metadata tables
• a Bun-first CLI for planning migrations, applying them, and emitting schema artifacts

Install

bun add @sedrino/db-schema

The CLI is Bun-first. If you want to run sedrino-db, make sure bun is available on PATH.

Example

import { compileSchemaToDrizzle, createMigration, planMigration } from "@sedrino/db-schema";

const migration = createMigration(
  {
    id: "2026-04-08-001-create-account",
    name: "Create account table",
  },
  (m) => {
    m.createTable("account", (t) => {
      t.id("accountId", { prefix: "acct" });
      t.string("name").required();
      t.temporalInstant("createdAt").required().defaultNow();
    });
  },
);

const plan = planMigration({ migration });
const drizzleSource = compileSchemaToDrizzle(plan.nextSchema);

Supported migration operations

• create, drop, and rename tables
• add, drop, rename, and alter fields
• add and drop indexes
• add and drop unique indexes

The builder also supports higher-level relationship helpers:

• belongsTo("table", ...) for indexed foreign keys
• createJunctionTable(...) for many-to-many join tables with composite uniqueness and inferred through(...) relations

For SQLite safety, field adds, drops, and supported field alterations are emitted as table rebuilds. Unsafe cases still produce planner warnings and migrate apply will refuse to run them.

When a rebuild needs help populating data, the preferred API is higher-level transform helpers:

import { transforms } from "@sedrino/db-schema";

m.alterTable("account", (t) => {
  t.string("slug")
    .required()
    .backfill(transforms.slugFrom("name"));

  t.alterField("createdAt", (f) => {
    f.temporalInstant().using(transforms.epochMsFromIsoString("createdAt"));
  });
});

Raw backfillSql(...) and usingSql(...) are still available as escape hatches.

Relationship helpers look like:

m.createTable("contact", (t) => {
  t.id("contactId", { prefix: "ct" });
  t.belongsTo("account", {
    required: true,
    onDelete: "cascade",
  });
});

m.createJunctionTable("userGroupMembership", {
  left: { table: "user" },
  right: { table: "group" },
});

Typed JSON fields are also supported. The second json(...) argument is a TypeScript type string that is carried into generated Drizzle code via .$type<...>():

m.createTable("account", (t) => {
  t.id("accountId", { prefix: "acct" });
  t.string("name").required();
  t.json("metadata", "{ source?: string; score?: number; tags?: string[] }");
});

Docs

• docs/index.md
• docs/schema-document.md
• docs/migrations.md
• docs/planning-and-apply.md
• docs/expressions-and-transforms.md
• docs/relations.md
• docs/cli.md

CLI

sedrino-db migrate create create-account --dir db
sedrino-db migrate plan --dir db
sedrino-db migrate apply --dir db --url file:./local.db
sedrino-db migrate validate --dir db
sedrino-db migrate status --dir db --url file:./local.db
sedrino-db schema print --dir db
sedrino-db schema drizzle --dir db --out db/schema/schema.generated.ts