A CRDT framework with a powerful abstraction of shared data

Yjs is a CRDT implementation that exposes its internal data structure as shared types. Shared types are common data types like Map or Array with superpowers: changes are automatically distributed to other peers and merged without merge conflicts.

Yjs is network agnostic (p2p!), supports many existing rich text editors, offline editing, version snapshots, undo/redo and shared cursors. It scales well with an unlimited number of users and is well suited for even large documents.

• Demos: https://github.com/yjs/yjs-demos
• Discuss: https://discuss.yjs.dev
• Chat: Gitter | Discord
• Benchmark Yjs vs. Automerge: https://github.com/dmonad/crdt-benchmarks
• Podcast "Yjs Deep Dive into real time collaborative editing solutions":
• Podcast "Google Docs-style editing in Gutenberg with the YJS framework":

👷‍♀️ If you are looking for professional support, please consider supporting this project via a "support contract" on GitHub Sponsors. I will attend your issues quicker and we can discuss questions and problems in regular video conferences. Otherwise you can find help on our community discussion board.

Please contribute to the project financially - especially if your company relies on Yjs.

• Support Contract with the Maintainer - By contributing financially to the open-source Yjs project, you can receive professional support directly from the author. This includes the opportunity for weekly video calls to discuss your specific challenges.
• Synergy Codes - Specializing in consulting and developing real-time collaborative editing solutions for visual apps, Synergy Codes focuses on interactive diagrams, complex graphs, charts, and various data visualization types. Their expertise empowers developers to build engaging and interactive visual experiences leveraging the power of Yjs. See their work in action at Visual Collaboration Showcase.

• AFFiNE A local-first, privacy-first, open source knowledge base. 🌟
• Huly - Open Source All-in-One Project Management Platform 🌟
• Cargo Site builder for designers and artists 🌟
• Gitbook Knowledge management for technical teams 🌟
• Evernote Note-taking app 🌟
• Lessonspace Enterprise platform for virtual classrooms and online training 🌟
• Ellipsus - Collaborative writing app for storytelling etc. Supports versioning, change attribution, and "blame". A solution for the whole publishing process (also selling) ⭐
• Dynaboard Build web apps collaboratively. ⭐
• Relm A collaborative gameworld for teamwork and community. ⭐
• Room.sh A meeting application with integrated collaborative drawing, editing, and coding tools. ⭐
• Nimbus Note A note-taking app designed by Nimbus Web. ⭐
• Pluxbox RadioManager A web-based app to collaboratively organize radio broadcasts. ⭐
• modyfi - Modyfi is the design platform built for multidisciplinary designers. Design, generate, animate, and more — without switching between apps. ⭐
• Sana A learning platform with collaborative text editing powered by Yjs.
• Serenity Notes End-to-end encrypted collaborative notes app.
• PRSM Collaborative mind-mapping and system visualisation. (source)
• Alldone A next-gen project management and collaboration platform.
• Living Spec A modern way for product teams to collaborate.
• Slidebeamer Presentation app.
• BlockSurvey End-to-end encryption for your forms/surveys.
• Skiff Private, decentralized workspace.
• JupyterLab Collaborative computational Notebooks
• JupyterCad Extension to JupyterLab that enables collaborative editing of 3d FreeCAD Models.
• Hyperquery A collaborative data workspace for sharing analyses, documentation, spreadsheets, and dashboards.
• Nosgestesclimat The french carbon footprint calculator has a group P2P mode based on yjs
• oorja.io Online meeting spaces extensible with collaborative apps, end-to-end encrypted.
• LegendKeeper Collaborative campaign planner and worldbuilding app for tabletop RPGs.
• IllumiDesk Build courses and content with A.I.
• btw Open-source Medium alternative
• AWS SageMaker Tools for building Machine Learning Models
• linear Streamline issues, projects, and product roadmaps.
• btw - Personal website builder
• AWS SageMaker - Machine Learning Service
• Arkiter - Live interview software
• Appflowy - They use Yrs
• Multi.app - Multiplayer app sharing: Point, draw and edit in shared apps as if they're on your computer. They are using Yrs.
• AppMaster A No-Code platform for creating production-ready applications with source code generation.
• Synthesia - Collaborative Video Editor
• thinkdeli - A fast and simple notes app powered by AI
• ourboard - A collaborative whiteboard applicaiton
• Ellie.ai - Data Product Design and Collaboration
• GoPeer - Collaborative tutoring
• screen.garden - Collaborative backend for PKM apps.
• NextCloud - Content Collaboration Platform
• keystatic - git-based CMS
• QDAcity - Collaborative qualitative data analysis platform
• Kanbert - Project management software
• Eclipse Theia - A cloud & desktop IDE that runs in the browser.
• ScienHub - Collaborative LaTeX editor in the browser.

• Open Collaboration Tools - Collaborative editing for your IDE or custom editor

• Overview
  • Bindings
  • Providers
  • Tooling
  • Ports
• Getting Started
• API
  • Shared Types
  • Y.Doc
  • Document Updates
  • Relative Positions
  • Y.UndoManager
• Yjs CRDT Algorithm
• License and Author

This repository contains a collection of shared types that can be observed for changes and manipulated concurrently. Network functionality and two-way-bindings are implemented in separate modules.

Setting up the communication between clients, managing awareness information, and storing shared data for offline usage is quite a hassle. Providers manage all that for you and are the perfect starting point for your collaborative app.

This list of providers is incomplete. Please open PRs to add your providers to this list!

y-websocket

A module that contains a simple websocket backend and a websocket client that connects to that backend. y-redis, y-sweet, ypy-websocket and Hocuspocus (see below) are alternative backends to y-websocket.

y-webrtc

Propagates document updates peer-to-peer using WebRTC. The peers exchange signaling data over signaling servers. Publically available signaling servers are available. Communication over the signaling servers can be encrypted by providing a shared secret, keeping the connection information and the shared document private.

@liveblocks/yjs 🌟

Liveblocks Yjs provides a fully hosted WebSocket infrastructure and persisted data store for Yjs documents. No configuration or maintenance is required. It also features Yjs webhook events, REST API to read and update Yjs documents, and a browser DevTools extension.

y-sweet ⭐

A standalone yjs server with persistence to S3 or filesystem. They offer a cloud service as well.

Hocuspocus ⭐

A standalone extensible yjs server with sqlite persistence, webhooks, auth and more.

@superviz/yjs

SuperViz Yjs Provider comes with a secure, scalable real-time infrastructure for Yjs documents, fully compatible with a set of real-time collaboration components offered by SuperViz. This solution ensures synchronization, offline editing, and real-time updates, enabling multiple users to collaborate effectively within shared workspaces.

PartyKit

Cloud service for building multiplayer apps.

y-libp2p

Uses libp2p to propagate updates via GossipSub. Also includes a peer-sync mechanism to catch up on missed updates.

y-dat

[WIP] Write document updates efficiently to the dat network using multifeed. Each client has an append-only log of CRDT local updates (hypercore). Multifeed manages and sync hypercores and y-dat listens to changes and applies them to the Yjs document.

Matrix-CRDT

Use Matrix as an off-the-shelf backend for Yjs by using the MatrixProvider. Use Matrix as transport and storage of Yjs updates, so you can focus building your client app and Matrix can provide powerful features like Authentication, Authorization, Federation, hosting (self-hosting or SaaS) and even End-to-End Encryption (E2EE).

yrb-actioncable

An ActionCable companion for Yjs clients. There is a fitting redis extension as well.

ypy-websocket

Websocket backend, written in Python.

Tinybase

The reactive data store for local-first apps. They support multiple CRDTs and different network technologies.

y-webxdc

Provider for sharing data in webxdc chat apps.

secsync

An architecture to relay end-to-end encrypted CRDTs over a central service.

y-indexeddb

Efficiently persists document updates to the browsers indexeddb database. The document is immediately available and only diffs need to be synced through the network provider.

y-mongodb-provider

Adds persistent storage to a server with MongoDB. Can be used with the y-websocket provider.

@toeverything/y-indexeddb

Like y-indexeddb, but with sub-documents support and fully TypeScript.

y-fire

A database and connection provider for Yjs based on Firestore.

y-op-sqlite

Persist YJS updates in your React Native app using op-sqlite , the fastest SQLite library for React Native.

y-postgresql

Provides persistent storage for a web server using PostgreSQL and is easily compatible with y-websocket.

• y-sweet debugger
• liveblocks devtools
• Yjs inspector

There are several Yjs-compatible ports to other programming languages.

• y-octo - Rust implementation by AFFiNE
• y-crdt - Rust implementation with multiple language bindings to other languages
  • yrs - Rust interface
  • ypy - Python binding
  • yrb - Ruby binding
  • yswift - Swift binding
  • yffi - C-FFI
  • ywasm - WASM binding
  • y_ex - Elixir bindings
• ycs - .Net compatible C# implementation.

Install Yjs and a provider with your favorite package manager:

npm i yjs y-websocket

Start the y-websocket server:

PORT=1234 node ./node_modules/y-websocket/bin/server.cjs

import * as Y from 'yjs';

const doc = new Y.Doc();
const yarray = doc.getArray('my-array')
yarray.observe(event => {
  console.log('yarray was modified')
})
// every time a local or remote client modifies yarray, the observer is called
yarray.insert(0, ['val']) // => "yarray was modified"

Remember, shared types are just plain old data types. The only limitation is that a shared type must exist only once in the shared document.

const ymap = doc.getMap('map')
const foodArray = new Y.Array()
foodArray.insert(0, ['apple', 'banana'])
ymap.set('food', foodArray)
ymap.get('food') === foodArray // => true
ymap.set('fruit', foodArray) // => Error! foodArray is already defined

Now you understand how types are defined on a shared document. Next you can jump to the demo repository or continue reading the API docs.

Any of the Yjs providers can be combined with each other. So you can sync data over different network technologies.

In most cases you want to use a network provider (like y-websocket or y-webrtc) in combination with a persistence provider (y-indexeddb in the browser). Persistence allows you to load the document faster and to persist data that is created while offline.

For the sake of this demo we combine two different network providers with a persistence provider.

import * as Y from 'yjs'
import { WebrtcProvider } from 'y-webrtc'
import { WebsocketProvider } from 'y-websocket'
import { IndexeddbPersistence } from 'y-indexeddb'

const ydoc = new Y.Doc()

// this allows you to instantly get the (cached) documents data
const indexeddbProvider = new IndexeddbPersistence('count-demo', ydoc)
indexeddbProvider.whenSynced.then(() => {
  console.log('loaded data from indexed db')
})

// Sync clients with the y-webrtc provider.
const webrtcProvider = new WebrtcProvider('count-demo', ydoc)

// Sync clients with the y-websocket provider
const websocketProvider = new WebsocketProvider(
  'wss://demos.yjs.dev', 'count-demo', ydoc
)

// array of numbers which produce a sum
const yarray = ydoc.getArray('count')

// observe changes of the sum
yarray.observe(event => {
  // print updates when the data changes
  console.log('new sum: ' + yarray.toArray().reduce((a,b) => a + b))
})

// add 1 to the sum
yarray.push([1]) // => "new sum: 1"

import * as Y from 'yjs'

const yarray = new Y.Array()

const ymap = new Y.Map()

const ytext = new Y.Text()

const yxml = new Y.XmlFragment()

const yxml = new Y.XmlElement()

const doc = new Y.Doc()

clientID

A unique id that identifies this client. (readonly)

gc

Whether garbage collection is enabled on this doc instance. Set `doc.gc = false` in order to disable gc and be able to restore old content. See https://github.com/yjs/yjs#yjs-crdt-algorithm for more information about gc in Yjs.

transact(function(Transaction):void [, origin:any])

Every change on the shared document happens in a transaction. Observer calls and the update event are called after each transaction. You should bundle changes into a single transaction to reduce the amount of event calls. I.e. doc.transact(() => { yarray.insert(..); ymap.set(..) }) triggers a single change event.
You can specify an optional origin parameter that is stored on transaction.origin and on('update', (update, origin) => ..).

toJSON():any

Deprecated: It is recommended to call toJSON directly on the shared types. Converts the entire document into a js object, recursively traversing each yjs type. Doesn't log types that have not been defined (using ydoc.getType(..)).

get(string, Y.[TypeClass]):[Type]

Define a shared type.

getArray(string):Y.Array

Define a shared Y.Array type. Is equivalent to y.get(string, Y.Array).

getMap(string):Y.Map

Define a shared Y.Map type. Is equivalent to y.get(string, Y.Map).

getText(string):Y.Text

Define a shared Y.Text type. Is equivalent to y.get(string, Y.Text).

getXmlElement(string, string):Y.XmlElement

Define a shared Y.XmlElement type. Is equivalent to y.get(string, Y.XmlElement).

getXmlFragment(string):Y.XmlFragment

Define a shared Y.XmlFragment type. Is equivalent to y.get(string, Y.XmlFragment).

on(string, function)

Register an event listener on the shared type

off(string, function)

Unregister an event listener from the shared type

on('update', function(updateMessage:Uint8Array, origin:any, Y.Doc):void)

Listen to document updates. Document updates must be transmitted to all other peers. You can apply document updates in any order and multiple times. Use `updateV2` to receive V2 events.

on('beforeTransaction', function(Y.Transaction, Y.Doc):void)

Emitted before each transaction.

on('afterTransaction', function(Y.Transaction, Y.Doc):void)

Emitted after each transaction.

on('beforeAllTransactions', function(Y.Doc):void)

Transactions can be nested (e.g. when an event within a transaction calls another transaction). Emitted before the first transaction.

on('afterAllTransactions', function(Y.Doc, Array<Y.Transaction>):void)

Emitted after the last transaction is cleaned up.

Changes on the shared document are encoded into document updates. Document updates are commutative and idempotent. This means that they can be applied in any order and multiple times.

const doc1 = new Y.Doc()
const doc2 = new Y.Doc()

doc1.on('update', update => {
  Y.applyUpdate(doc2, update)
})

doc2.on('update', update => {
  Y.applyUpdate(doc1, update)
})

// All changes are also applied to the other document
doc1.getArray('myarray').insert(0, ['Hello doc2, you got this?'])
doc2.getArray('myarray').get(0) // => 'Hello doc2, you got this?'

Yjs internally maintains a state vector that denotes the next expected clock from each client. In a different interpretation it holds the number of structs created by each client. When two clients sync, you can either exchange the complete document structure or only the differences by sending the state vector to compute the differences.

const state1 = Y.encodeStateAsUpdate(ydoc1)
const state2 = Y.encodeStateAsUpdate(ydoc2)
Y.applyUpdate(ydoc1, state2)
Y.applyUpdate(ydoc2, state1)

This example shows how to sync two clients with the minimal amount of exchanged data by computing only the differences using the state vector of the remote client. Syncing clients using the state vector requires another roundtrip, but can save a lot of bandwidth.

const stateVector1 = Y.encodeStateVector(ydoc1)
const stateVector2 = Y.encodeStateVector(ydoc2)
const diff1 = Y.encodeStateAsUpdate(ydoc1, stateVector2)
const diff2 = Y.encodeStateAsUpdate(ydoc2, stateVector1)
Y.applyUpdate(ydoc1, diff2)
Y.applyUpdate(ydoc2, diff1)

It is possible to sync clients and compute delta updates without loading the Yjs document to memory. Yjs exposes an API to compute the differences directly on the binary document updates.

// encode the current state as a binary buffer
let currentState1 = Y.encodeStateAsUpdate(ydoc1)
let currentState2 = Y.encodeStateAsUpdate(ydoc2)
// now we can continue syncing clients using state vectors without using the Y.Doc
ydoc1.destroy()
ydoc2.destroy()

const stateVector1 = Y.encodeStateVectorFromUpdate(currentState1)
const stateVector2 = Y.encodeStateVectorFromUpdate(currentState2)
const diff1 = Y.diffUpdate(currentState1, stateVector2)
const diff2 = Y.diffUpdate(currentState2, stateVector1)

// sync clients
currentState1 = Y.mergeUpdates([currentState1, diff2])
currentState2 = Y.mergeUpdates([currentState2, diff1])

If one of your users runs into a weird bug (e.g. the rich-text editor throws error messages), then you don't have to request the full document from your user. Instead, they can obfuscate the document (i.e. replace the content with meaningless generated content) before sending it to you. Note that someone might still deduce the type of content by looking at the general structure of the document. But this is much better than requesting the original document.

Obfuscated updates contain all the CRDT-related data that is required for merging. So it is safe to merge obfuscated updates.

const ydoc = new Y.Doc()
// perform some changes..
ydoc.getText().insert(0, 'hello world')
const update = Y.encodeStateAsUpdate(ydoc)
// the below update contains scrambled data
const obfuscatedUpdate = Y.obfuscateUpdate(update)
const ydoc2 = new Y.Doc()
Y.applyUpdate(ydoc2, obfuscatedUpdate)
ydoc2.getText().toString() // => "00000000000"

Yjs implements two update formats. By default you are using the V1 update format. You can opt-in into the V2 update format which provides much better compression. It is not yet used by all providers. However, you can already use it if you are building your own provider. All below functions are available with the suffix "V2". E.g. Y.applyUpdate ⇒ Y.applyUpdateV2. Also when listening to updates you need to specifically need listen for V2 events e.g. yDoc.on('updateV2', …). We also support conversion functions between both formats: Y.convertUpdateFormatV1ToV2 & Y.convertUpdateFormatV2ToV1.

Y.applyUpdate(Y.Doc, update:Uint8Array, [transactionOrigin:any])

Apply a document update on the shared document. Optionally you can specify transactionOrigin that will be stored on transaction.origin and ydoc.on('update', (update, origin) => ..).

Y.encodeStateAsUpdate(Y.Doc, [encodedTargetStateVector:Uint8Array]):Uint8Array

Encode the document state as a single update message that can be applied on the remote document. Optionally specify the target state vector to only write the differences to the update message.

Y.encodeStateVector(Y.Doc):Uint8Array

Computes the state vector and encodes it into an Uint8Array.

Y.mergeUpdates(Array<Uint8Array>)

Merge several document updates into a single document update while removing duplicate information. The merged document update is always smaller than the separate updates because of the compressed encoding.

Y.encodeStateVectorFromUpdate(Uint8Array): Uint8Array

Computes the state vector from a document update and encodes it into an Uint8Array.

Y.diffUpdate(update: Uint8Array, stateVector: Uint8Array): Uint8Array

Encode the missing differences to another update message. This function works similarly to Y.encodeStateAsUpdate(ydoc, stateVector) but works on updates instead.

convertUpdateFormatV1ToV2

Convert V1 update format to the V2 update format.

convertUpdateFormatV2ToV1

Convert V2 update format to the V1 update format.

When working with collaborative documents, we often need to work with positions. Positions may represent cursor locations, selection ranges, or even assign a comment to a range of text. Normal index-positions (expressed as integers) are not convenient to use because the index-range is invalidated as soon as a remote change manipulates the document. Relative positions give you a powerful API to express positions.

A relative position is fixated to an element in the shared document and is not affected by remote changes. I.e. given the document "a|c", the relative position is attached to c. When a remote user modifies the document by inserting a character before the cursor, the cursor will stay attached to the character c. insert(1, 'x')("a|c") = "ax|c". When the relative position is set to the end of the document, it will stay attached to the end of the document.

const relPos = Y.createRelativePositionFromTypeIndex(ytext, 2)
const pos = Y.createAbsolutePositionFromRelativePosition(relPos, doc)
pos.type === ytext // => true
pos.index === 2 // => true

const relPos = Y.createRelativePositionFromTypeIndex(ytext, 2)
const encodedRelPos = JSON.stringify(relPos)
// send encodedRelPos to remote client..
const parsedRelPos = JSON.parse(encodedRelPos)
const pos = Y.createAbsolutePositionFromRelativePosition(parsedRelPos, remoteDoc)
pos.type === remoteytext // => true
pos.index === 2 // => true

const relPos = Y.createRelativePositionFromTypeIndex(ytext, 2)
const encodedRelPos = Y.encodeRelativePosition(relPos)
// send encodedRelPos to remote client..
const parsedRelPos = Y.decodeRelativePosition(encodedRelPos)
const pos = Y.createAbsolutePositionFromRelativePosition(parsedRelPos, remoteDoc)
pos.type === remoteytext // => true
pos.index === 2 // => true

Y.createRelativePositionFromTypeIndex(type:Uint8Array|Y.Type, index: number [, assoc=0])

Create a relative position fixated to the i-th element in any sequence-like shared type (if assoc >= 0). By default, the position associates with the character that comes after the specified index position. If assoc < 0, then the relative position associates with the character before the specified index position.

Y.createAbsolutePositionFromRelativePosition(RelativePosition, Y.Doc): { type: Y.AbstractType, index: number, assoc: number } | null

Create an absolute position from a relative position. If the relative position cannot be referenced, or the type is deleted, then the result is null.

Y.encodeRelativePosition(RelativePosition):Uint8Array

Encode a relative position to an Uint8Array. Binary data is the preferred encoding format for document updates. If you prefer JSON encoding, you can simply JSON.stringify / JSON.parse the relative position instead.

Y.decodeRelativePosition(Uint8Array):RelativePosition

Decode a binary-encoded relative position to a RelativePositon object.

Yjs ships with an Undo/Redo manager for selective undo/redo of changes on a Yjs type. The changes can be optionally scoped to transaction origins.

const ytext = doc.getText('text')
const undoManager = new Y.UndoManager(ytext)

ytext.insert(0, 'abc')
undoManager.undo()
ytext.toString() // => ''
undoManager.redo()
ytext.toString() // => 'abc'

constructor(scope:Y.AbstractType|Array<Y.AbstractType> [, {captureTimeout:number,trackedOrigins:Set<any>,deleteFilter:function(item):boolean}])

Accepts either single type as scope or an array of types.

undo()

redo()

stopCapturing()

on('stack-item-added', { stackItem: { meta: Map<any,any> }, type: 'undo' | 'redo' })

Register an event that is called when a StackItem is added to the undo- or the redo-stack.

on('stack-item-updated', { stackItem: { meta: Map<any,any> }, type: 'undo' | 'redo' })

Register an event that is called when an existing StackItem is updated. This happens when two changes happen within a "captureInterval".

on('stack-item-popped', { stackItem: { meta: Map<any,any> }, type: 'undo' | 'redo' })

Register an event that is called when a StackItem is popped from the undo- or the redo-stack.

on('stack-cleared', { undoStackCleared: boolean, redoStackCleared: boolean })

Register an event that is called when the undo- and/or the redo-stack is cleared.

UndoManager merges Undo-StackItems if they are created within time-gap smaller than options.captureTimeout. Call um.stopCapturing() so that the next StackItem won't be merged.

// without stopCapturing
ytext.insert(0, 'a')
ytext.insert(1, 'b')
undoManager.undo()
ytext.toString() // => '' (note that 'ab' was removed)
// with stopCapturing
ytext.insert(0, 'a')
undoManager.stopCapturing()
ytext.insert(0, 'b')
undoManager.undo()
ytext.toString() // => 'a' (note that only 'b' was removed)

Every change on the shared document has an origin. If no origin was specified, it defaults to null. By specifying trackedOrigins you can selectively specify which changes should be tracked by UndoManager. The UndoManager instance is always added to trackedOrigins.

class CustomBinding {}

const ytext = doc.getText('text')
const undoManager = new Y.UndoManager(ytext, {
  trackedOrigins: new Set([42, CustomBinding])
})

ytext.insert(0, 'abc')
undoManager.undo()
ytext.toString() // => 'abc' (does not track because origin `null` and not part
                 //           of `trackedTransactionOrigins`)
ytext.delete(0, 3) // revert change

doc.transact(() => {
  ytext.insert(0, 'abc')
}, 42)
undoManager.undo()
ytext.toString() // => '' (tracked because origin is an instance of `trackedTransactionorigins`)

doc.transact(() => {
  ytext.insert(0, 'abc')
}, 41)
undoManager.undo()
ytext.toString() // => 'abc' (not tracked because 41 is not an instance of
                 //        `trackedTransactionorigins`)
ytext.delete(0, 3) // revert change

doc.transact(() => {
  ytext.insert(0, 'abc')
}, new CustomBinding())
undoManager.undo()
ytext.toString() // => '' (tracked because origin is a `CustomBinding` and
                 //        `CustomBinding` is in `trackedTransactionorigins`)

When undoing or redoing a previous action, it is often expected to restore additional meta information like the cursor location or the view on the document. You can assign meta-information to Undo-/Redo-StackItems.

const ytext = doc.getText('text')
const undoManager = new Y.UndoManager(ytext, {
  trackedOrigins: new Set([42, CustomBinding])
})

undoManager.on('stack-item-added', event => {
  // save the current cursor location on the stack-item
  event.stackItem.meta.set('cursor-location', getRelativeCursorLocation())
})

undoManager.on('stack-item-popped', event => {
  // restore the current cursor location on the stack-item
  restoreCursorLocation(event.stackItem.meta.get('cursor-location'))
})

Conflict-free replicated data types (CRDT) for collaborative editing are an alternative approach to operational transformation (OT). A very simple differentiation between the two approaches is that OT attempts to transform index positions to ensure convergence (all clients end up with the same content), while CRDTs use mathematical models that usually do not involve index transformations, like linked lists. OT is currently the de-facto standard for shared editing on text. OT approaches that support shared editing without a central source of truth (a central server) require too much bookkeeping to be viable in practice. CRDTs are better suited for distributed systems, provide additional guarantees that the document can be synced with remote clients, and do not require a central source of truth.

Yjs implements a modified version of the algorithm described in this paper. This article explains a simple optimization on the CRDT model and gives more insight about the performance characteristics in Yjs. More information about the specific implementation is available in INTERNALS.md and in this walkthrough of the Yjs codebase.

CRDTs that are suitable for shared text editing suffer from the fact that they only grow in size. There are CRDTs that do not grow in size, but they do not have the characteristics that are benificial for shared text editing (like intention preservation). Yjs implements many improvements to the original algorithm that diminish the trade-off that the document only grows in size. We can't garbage collect deleted structs (tombstones) while ensuring a unique order of the structs. But we can 1. merge preceeding structs into a single struct to reduce the amount of meta information, 2. we can delete content from the struct if it is deleted, and 3. we can garbage collect tombstones if we don't care about the order of the structs anymore (e.g. if the parent was deleted).

Examples:

1. If a user inserts elements in sequence, the struct will be merged into a single struct. E.g. text.insert(0, 'a'), text.insert(1, 'b'); is first represented as two structs ([{id: {client, clock: 0}, content: 'a'}, {id: {client, clock: 1}, content: 'b'}) and then merged into a single struct: [{id: {client, clock: 0}, content: 'ab'}].
2. When a struct that contains content (e.g. ItemString) is deleted, the struct will be replaced with an ItemDeleted that does not contain content anymore.
3. When a type is deleted, all child elements are transformed to GC structs. A GC struct only denotes the existence of a struct and that it is deleted. GC structs can always be merged with other GC structs if the id's are adjacent.

Especially when working on structured content (e.g. shared editing on ProseMirror), these improvements yield very good results when benchmarking random document edits. In practice they show even better results, because users usually edit text in sequence, resulting in structs that can easily be merged. The benchmarks show that even in the worst case scenario that a user edits text from right to left, Yjs achieves good performance even for huge documents.

Yjs has the ability to exchange only the differences when syncing two clients. We use lamport timestamps to identify structs and to track in which order a client created them. Each struct has an struct.id = { client: number, clock: number} that uniquely identifies a struct. We define the next expected clock by each client as the state vector. This data structure is similar to the version vectors data structure. But we use state vectors only to describe the state of the local document, so we can compute the missing struct of the remote client. We do not use it to track causality.

Yjs and all related projects are MIT licensed.

Yjs is based on my research as a student at the RWTH i5. Now I am working on Yjs in my spare time.

Fund this project by donating on GitHub Sponsors or hiring me as a contractor for your collaborative app.