Caracara: React + Dat + Automerge

How to create a simple collaborative editor on the distributed web

This article was originally written as a GEUT's tech insight document. You can find the original post together with the code here: https://github.com/geut/caracara

🔗 dat://caracara.hashbase.io/

Introduction: The idea #

The goal is to combine, again, Dat with some web development. The result will be a new experimental app, a mix between two kingdoms. When you combine something like this (at first two different worlds into one) the result does not mean to be the sum of all the benefits or even cons: a complete new thing. Based on this, what can you expect from a web app combined with Dat? You can see it for yourself.

In particular, when we say web app we are talking about PWAs where offline support is something that comes first and although it is not a key feature, it is common in the decentralized world. The addition of Dat plays a part in the efficient spreading of the content among peers.

All in all, we at GEUT believe that it is worth the effort of exploring these combinations.

Start #

Remember, our goal is to create a simple collaborative text editor. This means, peer A writes some text, shares the copy with other peer, call it B, automatically authorizing B to be a new writer.

OK, let's start with the bootstrap of our web app. We are going to be using React for this matter. More specifically, create-react-app to make things more quickly.

npx create-react-app caracara

After that,

cd caracara && npm start

Great, the app is running locally now. We are going to add a new module called @geut/saga. This will give P2P "powers" to our app. ✨

Saga #

Saga is a helper module developed by GEUT, that you can use to easily share operations between peers. You can think about this, like applications chatting each other. On top of this, we can do whatever we want with the operations. In this case, the meaning of the operation will be to regenerate the document history in every peer. Since saga will help us to establish a connection and exchange messages with other peers we need to see how it works.

First, install the dependencies:

npm install @geut/saga @geut/discovery-swarm-webrtc

Now we need to create a connection, this will allow us to talk with a swarm of peers:

This is an extract of swarm.js

// saga requires:
// - a [random-access-storage](https://github.com/random-access-storage) compatible module
// - a public key
// - and something that "identifies" the new Dat document
const saga = Saga(ram, publicKey, username);

// The initialization of saga is async. We are making sure hyperdb is ready.
await saga.initialize();

// Then, we create the swarm
const sw = swarm({
id: username,
stream: () => {
return saga.replicate();

// Now we need to say that this peer want to joins the swarm.
sw.join(signalhub(discoveryKey, signalUrls), webrtcOpts);
// Have in mind we are using webrtc to connect with other peers.
// Webrtc requires a signalhub that is going to be used to discover other peers.
// It's only used at first and then the communication is between peers.
// Nevertheless, is a central point in our decentralized solution.
// We are using the Dat discoveryKey for this purpose.

sw.on('connection', async peer => {
try {
// Finally, when we connect with other peer we let saga knows about it
await saga.connect(peer);
} catch (err) {

These are the important bits of the swarm creation. Now, let's see how we use this inside our app.

const SwarmContext = React.createContext(null);

class SwarmProvider extends Component {
state = {
swarmReady: false

async componentDidMount() {
const { username, match: { params: { draftId = null } } = {} } = this.props;
const swarm = await initComm(username, draftId);

swarmReady: true,
hasDraftId: !!draftId

render() {
const { swarmReady, swarm, hasDraftId } = this.state;
const { children } = this.props;
return swarmReady ? (
<SwarmContext.Provider value=>
) : null;
const withSwarm = WrappedComponent => {
return class extends Component {
static displayName = `WithSwarm${WrappedComponent.displayName}`;

render() {
return (
{({ swarm, hasDraftId }) => (


As you can see we have created a provider using the React Context API and also a withSwarm HOC for easy consuming using idiomatic React.

The editor #

Great! We've just achieved the swarm creation phase and we have also shown how to connect saga with our app. Now it's time to start playing a bit with the editor.

There are two new important dependencies at this step:

npm install automerge diff-match-patch

For this part we are going to highlight the Doc.js and Editor.js components.

componentDidMount() {
const { swarm } = this.props;
if (!swarm) return;

if (this.state.attachedEvents) return;

swarm.on('join', data => {
console.log('NEW COLLABORATOR', data);
const { username } = data;
this.setState(({ collaborators }) => ({
collaborators: new Set(collaborators).add(username)

swarm.on('operation', data => {
const { username, operation } = data;
if (username === this.props.username) return;
console.log('INCOMING OPERATION');
console.log('new operation from:', username);
console.log('new operation content:', operation);
if (operation.peerValue.length === 0) return;
let newDoc;
if (!this.doc) {
newDoc = Automerge.applyChanges(Automerge.init(), operation.peerValue);
} else {
newDoc = Automerge.applyChanges(this.doc, operation.peerValue); // peerValue are automerge changes
if (operation.original) {
this.original = operation.original;
this.doc = newDoc;
this.setState(prevState => ({
text: newDoc.text.join(''),
localHistory: [...this.state.localHistory, operation.diff]

this.setState({ attachedEvents: true });

componentWillUnmount() {
// Remember to remove listeners on unmount!
const { swarm } = this.props;
if (swarm !== null) swarm.removeAllListeners('operation');

At component's mount we setup saga to be listening to new operations. In our case, operations are Automerge changes. As soon as an operation arrives, we apply it. (There is a special case which is the document creation, that's why we use Automerge.init(), instead of the previous state: this.doc)

Next, we have the updatePeerValue function. For the sake of simplicity we are not using anything extra to maintain state. So, we will be passing down this function to our child component: the Editor. The Editor maintains a simple textarea component, as you can imagine when the textarea changes, that will trigger our updatePeerValue function.

We will use two important libraries:

  • Automerge will handle document transformations. It will give us changes that we can apply later. Automerge helps us maintain a CRDT data structure, this is super useful when doing collaborative work. 🆒
  • diff-match-patch is an excellent library from Google that we are going to use to differentiate the type of changes between previous and current text state. This in conjunction with Automerge creates a pretty solid strategy for handling and distributing changes without conflicts over a network, a decentralized one in our case.

Now let's go through the updatePeerValue function because it does several things actually.

updatePeerValue = val => {
const { comm, username } = this.props;
const { text } = val; // this is what the user wrote

// This is a simple way to detect the document autor
if (!this.original && comm.db.local.secretKey) {
// It is super important to share the doc:creation change first and
// then all the progressive changes.
this.doc = Automerge.change(Automerge.init(), 'doc:creation', doc => {
doc.text = new Automerge.Text();
const creationChange = Automerge.getChanges(Automerge.init(), this.doc);
this.original = true;

// here we are sharing a new operation related to the document creation
// the last param: diff is used to share the history message
peerValue: creationChange,
original: this.original,
diff: [`${username} created the doc - ${stamp(new Date(Date.now()))}`]

// this.dmp is a instance of diff-match-patch
// the usage of diff-match-patch here is based on https://lorefnon.tech/2018/09/23/using-google-diff-match-patch-with-automerge-text/
// Remember, our underlying editor is a simple textarea, so we need to detect specific modifications that happens to the text

const diff = this.dmp.diff_main(this.state.text, text);
const patches = this.dmp.patch_make(this.state.text, diff);

const updateAutomerge = (baseDoc, message, opFn) => {
return Automerge.change(baseDoc, message, doc => {

patches.forEach(patch => {
let idx = patch.start1;
let historyMessage = '';
let newDoc;
patch.diffs.forEach(([operation, changeText]) => {
switch (operation) {
case 1: // Insertion
historyMessage = `${username} added some text (${changeText}) - ${stamp(
new Date(Date.now())
newDoc = updateAutomerge(this.doc, historyMessage, doc => {
doc.text.insertAt(idx, ...changeText.split(''));
idx += changeText.length;
case 0: // No Change
idx += changeText.length;
newDoc = null;
case -1: // Deletion
historyMessage = `${username} removed some text (${changeText}) - ${stamp(
new Date(Date.now())
newDoc = updateAutomerge(this.doc, historyMessage, doc => {
for (let i = 0; i < changeText.length; i++) {
return doc;
if (newDoc) {
// thanks to diff-match-patch we can apply the specific change (insert, delete) to Automerge Text.
// Now we get the changes so we can share it with our peers.
const changes = Automerge.getChanges(this.doc, newDoc);
this.doc = newDoc;
localHistory: [...this.state.localHistory, changes[0].message]
() => {
// NOTE(dk): here we are sharing changes we made locally with our peers.
peerValue: changes,
diff: changes[0].message

Last but not least, the editor, this is the simplest piece of the whole project.

Here are the important bits.

onChange = e => {
const { value, selectionStart } = e.target;
() => {
if (this.state.selectionStart !== -1) {
this.setCaretToPos(this.taRef, this.state.selectionStart);
this.props.updatePeerValue({ text: value });
// Note(dk): a deboung will be better to not overload network
// but is making things difficult with the UX.
// Im going to revisit this strategy later, it requires a big refactor.
// this.debouncedPeerValue({ text: value });

The editor renders a textarea component with an onChange function attached to it. The onChange function updates the local value and call the updatePeerValue function that we have just seen.

Caveats #

The app is not totally decentralized. It is good to know this from the very beginning. We need to use some signaling servers to bootstrap the discovery of peers. From that point onwards, the communication is among peers. Another centralized moment is the first download of the app. This can be a github page, a CDN or whatever you want to use. Again, if the app works as a PWA with offline support, additional downloads will not be necessary.

Conclusions #

We have just dealt with the development of a Dat powered web app. Let's summarize the important parts:

  1. Install and instantiate saga. This is a top level interface on top of hyperdb.
  2. Create the swarm. This is useful to find another peers. Since we are creating a webapp we will be relying on WebRTC for peer to peer communication. A signal server will be required then. You can host your own if you want to. This is the one we recommend these days.
  3. After saga instantiation and swarm creation, we can start playing with our app. :tada:
    1. Since we were after a collaborative editor, two modules appear as a good choice here: Automerge and diff-match-patch.
    2. We will share operations that we get from Automerge changes.
    3. diff-match-patch is used to extract deltas from plain text.
  4. That's all! 👍

As you can see, thinking about sharing operations among peers opens the possibilities for new kind of applications and also makes things a bit easier. Looks like a win-win situation.

I hope this article helps you to start thinking on new possibilities and give it a try to use Dat on the web!

Thanks! 🦅

Acknowledgments #

I want to say thanks to the whole GEUT Team for their time and support. To Max (@elmasse) for sharing all his UX knowledge and React expertise to polish some rough edges and finally the Dat Project organization for their encouragement and openness. 🤗

📷 Caracara article photo by Florencia Lewis on Unsplash

If you want to contribute please create an issue with your suggestion and we can continue the discussion from there. 👍

Also, ⭐️ is ❤️ Well, not really ...but is cool. And you are cool. 😎