idb-ts

---

Introduction

idb-ts is a declarative, type-safe ORM layer for IndexedDB. Define your data models with TypeScript decorators, and the library handles schema creation, key generation, validation, querying, transactions, and data retention automatically - with no external runtime dependencies.

---

Installation

npm install idb-ts
pnpm add idb-ts
yarn add idb-ts
Copy

Requirement: reflect-metadata must be imported once at your application entry point, and experimentalDecorators and emitDecoratorMetadata must be enabled in your tsconfig.json.

{
  "compilerOptions": {
    "experimentalDecorators": true,
    "emitDecoratorMetadata": true
  }
}
Copy

---

Feature Overview

Feature	Description
Declarative entity definition	Define stores, keys, and indexes with class decorators
Full CRUD API	Create, read, update, delete, list, paginate, and count
Typed query builder	Chainable, type-checked filter, sort, and aggregation DSL
Key generation	Auto-increment, UUID v4, timestamp, random, or custom function
Composite keys	Multi-field primary keys for relational associations
Field validation	Per-property predicate rules enforced on write
Schema versioning	Automatic onupgradeneeded migration based on entity versions
Transaction API	Callback-based and explicit commit/rollback patterns
Data retention	Periodic background cleanup of expired records
Automatic timestamps	__idb_createdAt / __idb_updatedAt injected on every write

---

Quick Start

import 'reflect-metadata';
import { Database, DataClass, KeyPath, Index } from 'idb-ts';

@DataClass()
class User {
  @KeyPath({ generator: 'uuid' })
  id!: string;

  @Index({ unique: true })
  email!: string;

  name!: string;
  age!: number;
}

const db = await Database.build<{ User: EntityRepository<User> }>('mydb', [
  User,
]);

await db.User.create({
  id: '',
  name: 'Alice',
  age: 30,
  email: 'alice@example.com',
});
const alice = await db.User.findOneByIndex('email', 'alice@example.com');
Copy

---

Defining Entities

Every entity class must declare exactly one primary key field and be annotated with @DataClass(). Apply decorators in the order shown - TypeScript executes decorators bottom-up, so @DataClass must appear last (i.e., closest to the class keyword).

import { Database, DataClass, KeyPath, Index, Validate } from 'idb-ts';

@DataClass({ version: 1 })
class User {
  @KeyPath({ generator: 'uuid' })
  id!: string;

  @Index({ unique: true })
  @Validate(
    (v) => typeof v === 'string' && v.includes('@'),
    'must be a valid email',
  )
  email!: string;

  @Validate((v) => typeof v === 'number' && v >= 0, 'age must be non-negative')
  age!: number;

  name!: string;
}
Copy

Decorator reference

@DataClass(options?)

Marks a class as a managed entity. Must be applied exactly once per class, after all other idb-ts decorators.

Option	Type	Default	Description
version	number	1	Schema version. Increment when the entity's store or indexes change.

@KeyPath(options?)

Designates the decorated property as the primary key of the object store. Exactly one property per class may carry this decorator. For multi-field keys, use @CompositeKeyPath at the class level instead.

Option	Type	Default	Description
autoIncrement	boolean	false	Delegate key assignment to IndexedDB's auto-increment mechanism.
generator	'uuid' | 'timestamp' | 'random' | (item) => string | number	-	Automatic key generator invoked when the key field is absent or empty on create.

@CompositeKeyPath(fields, options?)

Class-level decorator for composite primary keys. Cannot be combined with @KeyPath. Write it below @DataClass (decorators are applied bottom-up, and the key path must be registered before @DataClass validates it).

Key generation is not supported for composite keys: passing generator or autoIncrement throws at decoration time. Provide every key field explicitly before create().

@DataClass()
@CompositeKeyPath(['userId', 'projectId'])
class UserProject {
  userId!: string;
  projectId!: string;
  role!: string;
}
Copy

@Index(options?)

Creates an IDB index on the decorated field, enabling efficient lookups via findByIndex and findOneByIndex.

Option	Type	Description
unique	boolean	Enforce uniqueness on the indexed field.

@Validate(predicate, message)

Attaches a validation rule to the decorated property. Rules are enforced on every create and update call. If any rule fails, the operation throws with a message listing all failing fields.

@Calculated(compute)

Derives the decorated property from the rest of the entity on every create and update. See Calculated Fields.

@RetentionPolicy(options)

Class-level decorator that configures automatic expiry and deletion of records. See Data Retention for full details.

---

Database Initialisation

const db = await Database.build<{
  User: EntityRepository<User>;
  Order: EntityRepository<Order>;
}>('shop', [User, Order]);
Copy

Database.build opens (or upgrades) the IDB database, reconciles the declared schema against the stored one (creating missing stores and indexes and removing indexes that are no longer declared), starts background retention jobs if applicable, and attaches typed repository properties to the returned object.

The declared database version is the highest version value across all registered entities; see Migration behaviour for how drift and downgrades are handled.

Inspecting database metadata

db.getDatabaseVersion(); // number - current IDB version
db.getEntityVersions(); // Map<string, number>
db.getEntityVersion('User'); // number | undefined
db.getAvailableEntities(); // string[]
Copy

Closing the connection

db.close(); // Stops the retention cleanup timer and closes the IDB connection.
Copy

---

CRUD Operations

Each entity is accessible as a named property on the database object. All methods return Promise.

// Create
await db.User.create(user);
await db.User.createMany([alice, bob, charlie]);

// Read
const user = await db.User.read('u1'); // by primary key
const page = await db.User.listPaginated(1, 20); // 1-based pagination
const all = await db.User.list();

// Update
await db.User.update(updatedUser);
await db.User.updateMany([user1, user2]);

// Delete
await db.User.delete('u1');
await db.User.deleteMany(['u1', 'u2']);
await db.User.deleteWhere((q) => q.where('age').lt(18));

// Utilities
const count = await db.User.count();
const exists = await db.User.exists('u1');
const keys = await db.User.getKeys(); // primary keys only, no record values
await db.User.clear();
Copy

Index lookups

const allAdmins = await db.User.findByIndex('role', 'admin');
const firstAdmin = await db.User.findOneByIndex('role', 'admin');
Copy

Querying a non-existent index throws immediately.

---

Automatic Timestamps

Every record written through a repository automatically receives two internal fields:

Field	Type	Set on
__idb_createdAt	number (ms since epoch)	create only
__idb_updatedAt	number (ms since epoch)	create and update

__idb_createdAt is preserved across updates; __idb_updatedAt is refreshed on every write.

const item = await db.Session.read(key);
console.log(item.__idb_createdAt, item.__idb_updatedAt);
Copy

---

Query Builder

EntityRepository.query() returns a typed QueryBuilder<T> for constructing complex filter expressions, sorting, pagination, and aggregations.

Filtering

const results = await db.User.query()
  .where('age')
  .gte(18)
  .and('status')
  .equals('active')
  .execute();
Copy

Available operators

Operator	Field types	Description
equals	any	Strict equality (===)
gt / gte / lt / lte	ComparableValue	Comparison
between(start, end)	ComparableValue	Inclusive range
notBetween(start, end)	ComparableValue	Outside range
startsWith / endsWith	string	Prefix / suffix match
contains	string | array	Substring or element membership
matches	string	Regular expression test
in(values) / notIn(values)	any	Membership test
containsAny(values)	array	At least one element matches
containsAll(values)	array	All elements present

TypeScript enforces operator/type compatibility at compile time - string-only operators are not exposed on numeric fields, and so on.

Logical grouping

// OR connector
const results = await db.User.query()
  .where('age')
  .gte(18)
  .or()
  .where('hasParentalConsent')
  .equals(true)
  .execute();

// Grouped sub-expression
const premiumOrTrial = await db.User.query()
  .where((qb) =>
    qb.where('type').equals('premium').and('status').equals('active'),
  )
  .or()
  .where('isTrial')
  .equals(true)
  .execute();
Copy

Sorting and pagination

await db.User.query()
  .where('status')
  .equals('active')
  .orderBy('createdAt', 'desc')
  .offset(20)
  .limit(10)
  .execute();
Copy

Reusing builders

A builder accumulates state: every where/orderBy/limit call mutates the same instance, so chaining more conditions onto an already-executed builder narrows it further. To derive variations from a shared base use clone(); to start over with the same instance use reset():

const adults = db.User.query().where('age').gte(18);

// Independent variations - neither affects the other or the base
const admins = await adults.clone().where('role').equals('admin').execute();
const guests = await adults.clone().where('role').equals('guest').execute();

// Reuse one instance from scratch
const query = db.User.query();
await query.where('role').equals('admin').execute();
await query.reset().where('age').lt(18).execute(); // fresh state
Copy

Index and range acceleration

When a field is indexed, you can constrain the initial IDB candidate set at the storage layer before in-memory filtering begins:

await db.Product.query().useIndex('price').range(10, 100).execute();
Copy

The two mechanisms are deliberately distinct:

• useIndex(...).range(start, end) narrows candidates natively at the IndexedDB layer via an IDBKeyRange — fast, but limited to one indexed field.
• .where(...) conditions are evaluated in memory after the candidates are fetched. They can target any field (including one different from the index), at the cost of scanning the fetched candidates.

Mixing them is valid and useful — the index range prunes the bulk, where() refines the rest. Calling range() without useIndex() throws at execution time instead of silently ignoring the bounds; express such bounds as where(field).between(start, end) instead.

Aggregations

await db.Order.query().where('status').equals('paid').count();
await db.Order.query().sum('amount');
await db.Order.query().avg('price');
await db.Order.query().min('createdAt');
await db.Order.query().max('createdAt');

// Grouped count
const byStatus = await db.Order.query().groupBy('status').count();
// [{ status: 'paid', count: 42 }, { status: 'pending', count: 7 }]
Copy

sum and avg are restricted to numeric fields. min and max accept any comparable field. groupBy(...).count() returns results sorted by group key.

---

Key Management

Auto-increment

@DataClass()
class Task {
  @KeyPath({ autoIncrement: true })
  id!: number; // Assigned by IndexedDB: 1, 2, 3, …

  title!: string;
}
Copy

Built-in generators

@DataClass()
class Document {
  @KeyPath({ generator: 'uuid' }) // RFC 4122 v4
  id!: string;
}

@DataClass()
class Event {
  @KeyPath({ generator: 'timestamp' }) // Date.now()
  id!: number;
}

@DataClass()
class Session {
  @KeyPath({ generator: 'random' }) // Base-36 random string
  id!: string;
}
Copy

Custom generator

@DataClass()
class Invoice {
  @KeyPath({
    generator: (entity) =>
      `INV-${entity.year}-${String(entity.number).padStart(4, '0')}`,
  })
  invoiceId!: string;

  year!: number;
  number!: number;
}
// invoiceId → "INV-2024-0001"
Copy

Using generators directly

import { KeyGenerators } from 'idb-ts';

KeyGenerators.uuid(); // "a1b2c3d4-..."
KeyGenerators.timestamp(); // 1696118400000
KeyGenerators.random(); // "xyz789abc"
Copy

Composite keys

Key generation (generator / autoIncrement) is not supported for composite keys and throws at decoration time.

@DataClass()
@CompositeKeyPath(['userId', 'projectId'])
class UserProject {
  userId!: string;
  projectId!: string;

  @Index()
  role!: string;

  joinedAt!: Date;
}

// Create
await db.UserProject.create(new UserProject('u1', 'p1', 'developer'));

// Read / update / delete with composite key tuple
const rel = await db.UserProject.read(['u1', 'p1']);
await db.UserProject.delete(['u1', 'p1']);
Copy

---

Field Validation

Validation rules are declared per-property with @Validate. All rules for an entity are evaluated before any write; a single thrown error enumerates every failing rule.

@DataClass()
class User {
  @KeyPath()
  id!: string;

  @Validate(
    (v) => typeof v === 'string' && /^[^\s@]+@[^\s@]+\.[^\s@]+$/.test(v),
    'must be a valid email address',
  )
  email!: string;

  @Validate(
    (v) => Number.isInteger(v) && v >= 0,
    'must be a non-negative integer',
  )
  age!: number;
}
Copy

Error format on failure:

Validation failed for User: email: must be a valid email address; age: must be a non-negative integer
Copy

---

Calculated Fields

@Calculated derives a property from the rest of the entity on every write. The compute function runs on create and update, before validation and timestamps:

import { Calculated } from 'idb-ts';

@DataClass()
class OrderLine {
  @KeyPath({ generator: 'uuid' })
  id!: string;

  quantity!: number;
  unitPrice!: number;

  @Calculated<OrderLine>((line) => line.quantity * line.unitPrice)
  total!: number;
}

await db.OrderLine.create({ id: '', quantity: 3, unitPrice: 9.5 } as OrderLine);
(await db.OrderLine.read(id))!.total; // 28.5 - computed and persisted
Copy

Semantics:

• Any value the caller assigns to a calculated field is overwritten by the compute function on write.
• @Validate rules on the same field see the computed value.
• The value is persisted, so it can be indexed with @Index and queried like any ordinary field.
• It reflects entity state as of the last write - update the inputs and the field recomputes on the next update.

---

Transactions

Callback form (recommended)

The callback receives a TransactionalDatabase handle. On successful return the transaction is committed automatically. Any thrown error triggers an automatic rollback before rethrowing.

await db.transaction(async (tx) => {
  await tx.User.create(user);
  await tx.Order.create(order);
  await tx.OrderItem.create(item);
});
Copy

Explicit form

const tx = await db.beginTransaction(['User', 'Order'], 'readwrite');
try {
  await tx.User.create(user);
  await tx.Order.create(order);
  await tx.commit();
} catch (error) {
  await tx.rollback();
  throw error;
}
Copy

Transaction semantics

All repository operations performed through the tx handle share the same native IDBTransaction, ensuring atomicity. beginTransaction accepts an array of entity names that determines the transaction scope; the callback form spans all registered entities. The default mode is 'readwrite'; pass 'readonly' for read-only workloads. Use tx.Entity.query() to run queries within the same transaction boundary.

---

Data Retention

@RetentionPolicy triggers a background cleanup job that deletes records whose age exceeds the configured threshold.

@RetentionPolicy({ seconds: 60 * 60 * 24 * 30 }) // 30-day retention
@DataClass()
class Session {
  @KeyPath({ generator: 'uuid' })
  id!: string;

  userId!: string;
}
Copy

Option	Type	Default	Description
seconds	number	-	(Required) Retention window in seconds. Must be a positive integer.
enabled	boolean	true	Set to false to suspend cleanup without removing the policy.
field	string	'__idb_createdAt'	Numeric timestamp field used to compute record age.

When multiple entities define retention policies, the cleanup interval is set to the GCD of all configured seconds values in milliseconds, so a single timer satisfies every policy efficiently. The job runs immediately on database open and then on each interval tick, using cursor-based readwrite transactions.

---

Schema Versioning

Increment an entity's version to trigger onupgradeneeded and update its object store on the user's next visit. The declared database version is the maximum across all registered entities, so adding a new high-version entity is sufficient to initiate a migration.

@DataClass({ version: 1 })
class User {
  /* ... */
}
@DataClass({ version: 2 })
class Post {
  /* ... */
}
@DataClass({ version: 3 })
class Comment {
  /* ... */
}

// Database opens at version 3 and reconciles the full declared schema
// (stores + indexes) inside the upgrade transaction.
const db = await Database.build('blog', [User, Post, Comment]);

console.log(db.getDatabaseVersion()); // 3
Copy

Migration behaviour

Migration is declarative: on every upgrade the actual IndexedDB schema is reconciled against the schema declared by your decorators.

Change	Handling
New entity / store	Created automatically.
Index added (even without a version bump)	Detected as schema drift after opening; the database is reopened one version higher and the index is created. Existing records are re-indexed by IndexedDB.
Index removed (even without a version bump)	Detected as drift; the stale index is deleted. Record data is not affected.
Entity version lowered	IndexedDB cannot downgrade. The database opens at the existing on-disk version instead of throwing VersionError; getDatabaseVersion() reports the on-disk version.
Key path / autoIncrement changed	Not applied — IndexedDB cannot change a store's key path in place. A warning is logged; migrate the data to a new entity or delete the database.
Entity no longer registered	Its store and data are preserved and a warning is logged. Re-register the entity to access the data again, or delete the store manually.

Because drift detection may reopen the database one version higher than declared, getDatabaseVersion() returns the actual IndexedDB version, which can exceed the highest entity version.

---

Bulk Operations

All repository bulk helpers iterate the corresponding single-item operation and therefore enforce validation and key generation per item. They are not issued as a single atomic transaction. For atomic batch writes, use the Transaction API.

await db.User.createMany([alice, bob, charlie]);
await db.User.updateMany([alice, bob]);
await db.User.deleteMany(['u1', 'u2', 'u3']);
Copy

---

Sync Adapters (experimental)

Bridge the local IndexedDB with any backend by implementing the two-method SyncAdapter interface. The library stays transport-agnostic - REST, WebSocket, or in-memory adapters all work the same way.

import type { SyncAdapter } from 'idb-ts';

class RestAdapter implements SyncAdapter {
  async push(entityName: string, records: unknown[]): Promise<void> {
    await fetch(`/api/sync/${entityName}`, {
      method: 'PUT',
      body: JSON.stringify(records),
    });
  }

  async pull(entityName: string): Promise<unknown[] | undefined> {
    const response = await fetch(`/api/sync/${entityName}`);
    return response.ok ? response.json() : undefined;
  }
}

const adapter = new RestAdapter();
await db.pushTo(adapter); // sends every entity's records to the adapter
await db.pullFrom(adapter); // upserts records returned by the adapter
Copy

• pushTo calls adapter.push(entityName, records) once per registered entity.
• pullFrom calls adapter.pull(entityName) per entity and upserts the returned records by primary key; returning undefined leaves that store untouched.
• Conflict resolution is intentionally left to the adapter/backend - locally, pulled records win by primary key.

---

Export / Import

Snapshot every registered store to a plain serialisable object, and load such a snapshot back - useful for backups, test fixtures, and moving data between environments.

// Export: { EntityName: records[] } for every registered entity
const dump = await db.exportDatabase();
localStorage.setItem('backup', JSON.stringify(dump));

// Import: writes records verbatim (timestamps preserved), upserting by key
await db.importDatabase(JSON.parse(localStorage.getItem('backup')!));

// Replace instead of merge
await db.importDatabase(dump, { clear: true });
Copy

• Records are written with put, so importing over existing keys overwrites those records; other records are kept unless clear: true is passed.
• Dump entries whose entity is not registered in this database are skipped.
• Internal __idb_createdAt / __idb_updatedAt fields survive the round-trip verbatim; validation and key generation are bypassed so the restored data matches the exported data exactly.

---

Performance

Already up to date Done in 383ms using pnpm v11.9.0

Suite 1: CRUD Operations

Operation	Ops	Total ms	Ops/s	Avg ms	P50	P95	P99	Min	Max
create (single)	200	11.668	17,141.483	0.058	0.044	0.103	0.13	0.039	0.883
read (by PK)	200	7.972	25,087.351	0.039	0.036	0.062	0.091	0.03	0.113
update (single)	200	133.17	1,501.843	0.665	0.615	1.01	1.184	0.545	2.036
findByIndex (email)	200	8.942	22,365.168	0.044	0.041	0.073	0.086	0.034	0.111
findOneByIndex (email)	200	10.246	19,519.677	0.051	0.044	0.057	0.072	0.039	0.998
count	200	8.28	24,155.266	0.041	0.04	0.049	0.055	0.037	0.079
exists	200	13.325	15,009.758	0.066	0.052	0.09	0.118	0.046	2.085
list (all)	50	96.56	517.814	1.931	1.771	4.279	6.013	1.515	6.013
listPaginated (1, 20)	200	398.616	501.736	1.993	1.765	3.022	8.919	1.51	10.708
query().where().gte().execute()	100	181.336	551.463	1.813	1.599	1.896	2.042	1.524	14.147
delete (single)	200	161.278	1,240.095	0.806	0.773	0.896	1.019	0.64	6.043

Suite 2: Batched CRUD (by batch size)

Operation	Ops	Total ms	Ops/s	Avg ms	P50	P95	P99	Min	Max
createMany (10)	3	1.421	2,111.564	0.473	0.513	0.528	0.528	0.377	0.528
read batch (10 keys)	3	0.5	5,998.488	0.166	0.16	0.189	0.189	0.151	0.189
updateMany (10)	3	4.472	670.902	1.49	1.452	1.646	1.646	1.37	1.646
deleteMany (10)	3	3.163	948.378	1.054	1.054	1.139	1.139	0.969	1.139
deleteWhere (10+ match)	3	0.841	3,566.321	0.28	0.278	0.312	0.312	0.249	0.312
createMany (50)	3	4.301	697.531	1.433	1.433	1.477	1.477	1.39	1.477
read batch (50 keys)	3	3.931	763.202	1.31	1.296	1.421	1.421	1.212	1.421
updateMany (50)	3	87.877	34.138	29.291	29.002	31.244	31.244	27.628	31.244
deleteMany (50)	3	93.633	32.04	31.21	31.53	33.2	33.2	28.899	33.2
deleteWhere (50+ match)	3	3.446	870.695	1.148	1.127	1.241	1.241	1.075	1.241
createMany (100)	3	9.571	313.443	3.19	3.074	3.8	3.8	2.695	3.8
read batch (100 keys)	3	12.769	234.949	4.255	4.077	5.665	5.665	3.023	5.665
updateMany (100)	3	314.564	9.537	104.853	104.907	104.963	104.963	104.69	104.963
deleteMany (100)	3	331.429	9.052	110.474	108.426	121.766	121.766	101.228	121.766
deleteWhere (100+ match)	3	6.954	431.435	2.317	2.254	2.481	2.481	2.216	2.481
createMany (500)	3	88.569	33.872	29.522	27.663	42.007	42.007	18.894	42.007
read batch (500 keys)	3	118.5	25.316	39.498	36.064	51.839	51.839	30.592	51.839
updateMany (500)	3	8,577.818	0.35	2,859.271	2,855.88	2,877.442	2,877.442	2,844.49	2,877.442
deleteMany (500)	3	9,452.42	0.317	3,150.804	3,097.544	3,450.784	3,450.784	2,904.084	3,450.784
deleteWhere (500+ match)	3	35.488	84.535	11.828	11.745	12.529	12.529	11.21	12.529

Suite 3: Mixed CRUD Operations

Operation	Ops	Total ms	Ops/s	Avg ms	P50	P95	P99	Min	Max
Read-heavy mix (70R/15U/10C/5D)	200	38.005	5,262.407	0.19	0.031	0.716	0.737	0.001	0.745
Write-heavy mix (20R/15U/50C/15D)	200	44.985	4,445.915	0.225	0.035	0.707	0.739	0.001	3.708
Mixed CRUD + queries	200	83.386	2,398.49	0.417	0.033	2.061	2.141	0.011	6.234
Cross-entity mix (User.read + Order.create)	200	6.295	31,770.944	0.031	0.024	0.058	0.076	0.018	0.095

Suite 4: Mixed Batched CRUD

Operation	Ops	Total ms	Ops/s	Avg ms	P50	P95	P99	Min	Max
Cycle: createMany -> readAll -> updateMany -> deleteMany (50)	5	104.266	47.954	20.852	15.473	36.333	36.333	9.223	36.333
createMany -> query filter -> deleteMany (50)	5	33.079	151.155	6.615	6.368	9.824	9.824	4.751	9.824
5 waves × createMany(50) + deleteMany(50)	3	292.584	10.253	97.527	94.694	108.554	108.554	89.333	108.554
Cross-entity batch: createMany(User) + createMany(Order) + deleteMany (×50)	3	136.071	22.047	45.356	43.887	49.364	49.364	42.816	49.364

Suite 5: Transaction Operations

Operation	Ops	Total ms	Ops/s	Avg ms	P50	P95	P99	Min	Max
tx: single create	100	7.5	13,332.676	0.075	0.063	0.121	0.209	0.06	0.37
tx: create 10 users	100	23.044	4,339.529	0.23	0.219	0.27	0.413	0.207	0.455
tx: read + update	100	194.689	513.64	1.947	1.906	1.986	3.09	1.836	5.522
tx: multi-entity create (User+Order+Session)	100	7.336	13,632.137	0.073	0.07	0.098	0.114	0.066	0.159
tx: 10 reads	100	15.108	6,618.828	0.151	0.141	0.176	0.272	0.135	0.364
tx: batch create 50 users	20	23.258	859.911	1.163	0.953	1.68	4.091	0.926	4.091
tx: query().where().gte()	100	946.761	105.623	9.467	8.473	18.988	19.246	8.265	19.705
tx (explicit): begin -> create -> commit	100	5.381	18,582.249	0.054	0.052	0.067	0.078	0.049	0.081

Suite 6: Mixed Transactions

Operation	Ops	Total ms	Ops/s	Avg ms	P50	P95	P99	Min	Max
tx mixed: read User -> create Order -> update User	100	53.926	1,854.389	0.539	0.437	0.462	0.59	0.41	10.318
tx mixed: query User + read Order + create Session	100	113.792	878.798	1.138	1.055	1.088	1.139	1.03	8.959
tx multi-entity: create User+Order+Session	100	8.889	11,250.085	0.089	0.087	0.102	0.106	0.082	0.132
tx batched: create 20 Users + 40 Orders + 20 Sessions	10	20.246	493.934	2.024	1.136	9.953	9.953	1.096	9.953
tx mixed: delete old orders -> create new orders	100	122.343	817.377	1.223	1.145	1.309	1.386	0.981	8.579
tx mixed: count Orders -> conditional create	100	7.079	14,125.357	0.071	0.069	0.083	0.099	0.065	0.103
tx complex: read User+Orders -> aggregate -> create Session	100	15.71	6,365.173	0.157	0.117	0.216	0.33	0.107	2.822

LLM / AI Assistant Instructions

Building with an AI coding assistant? Paste the block below into your assistant's context (system prompt, rules file, CLAUDE.md, .cursorrules, etc.) so it generates correct idb-ts code on the first try.

idb-ts cheat sheet (TypeScript ORM for IndexedDB, zero runtime deps):

Setup
- import 'reflect-metadata' once at the app entry point.
- tsconfig: "experimentalDecorators": true, "emitDecoratorMetadata": true.

Entities
- Decorate classes with @DataClass({ version?: number }).
- Exactly one primary key: @KeyPath({ autoIncrement?, generator? }) on a
  property, or @CompositeKeyPath(['fieldA','fieldB']) on the class
  (written BELOW @DataClass - decorators apply bottom-up).
- generator: 'uuid' | 'timestamp' | 'random' | (item) => string | number.
- Secondary indexes: @Index({ unique?: boolean }) on properties.
- Validation: @Validate((value, item) => boolean, 'message') on properties.
- Auto-expiry: @RetentionPolicy({ seconds, field?, enabled? }) on the class.

Database
- const db = await Database.build<{ User: EntityRepository<User> }>('name', [User]);
- Repositories are attached by class name: db.User, db.Order, ...
- db.close() when done. db.getDatabaseVersion(), db.getAvailableEntities().

Repository API (all Promise-based)
- create(item), createMany(items), read(key), update(item), updateMany(items)
- delete(key), deleteMany(keys), deleteWhere(q => q.where(...))
- list(), listPaginated(page, pageSize), count(), exists(key), clear()
- findByIndex(indexName, value), findOneByIndex(indexName, value)
- query() -> QueryBuilder

QueryBuilder (chainable)
- .where('field').equals/gt/gte/lt/lte/startsWith/endsWith/contains/
  matches/between/notBetween/in/notIn/containsAny/containsAll(...)
- .and('field')... / .or().where('field')...
- Nested groups: .where(qb => qb.where('a').equals(1).or().where('b').equals(2))
- .orderBy(field, 'asc'|'desc'), .limit(n), .offset(n)
- .useIndex(indexName).range(start, end) for IDB-level narrowing
- Terminals: .execute(), .count(), .sum(f), .avg(f), .min(f), .max(f),
  .groupBy(f).count()

Transactions
- await db.transaction(async tx => { await tx.User.create(u); ... })  // auto commit/rollback
- const tx = await db.beginTransaction(['User','Order']); await tx.commit() / tx.rollback()

Gotchas
- Store names are lower-cased class names; renaming a class = new store.
- Every write injects __idb_createdAt / __idb_updatedAt (ms timestamps).
- where() filters run in memory after candidates are fetched; use
  useIndex()+range() to narrow at the IndexedDB layer first.
- Composite keys are passed as arrays: db.UserProject.read(['u1','p1']).
Copy

For contributors (and their assistants)

Working on idb-ts itself: the entire library lives in index.ts; tests are in __tests__/ (Jest + fake-indexeddb). Key workflows: pnpm test (type check + Jest), pnpm lint, pnpm build (tsc + rollup into lib/, which is the only published artifact). Keep changes minimal, always add tests, and document user-facing behaviour in this README rather than separate files.

Useful Links

• GitHub: maifeeulasad/idb-ts
• NPM: idb-ts
• Demos: https://maifeeulasad.github.io/idb-ts/
• Live Editor: https://maifeeulasad.github.io/idb-ts/typescript/
• Code Coverage report: https://maifeeulasad.github.io/idb-ts/coverage/lcov-report/

🎉 Enjoy seamless IndexedDB integration with TypeScript! Happy coding! 🚀

Made by Maifee Ulasad with :heart: and :tea:. Licensed under MIT.