diff --git a/discojs-node/README.md b/discojs-node/README.md
new file mode 100644
index 000000000..a5c048ecb
--- /dev/null
+++ b/discojs-node/README.md
@@ -0,0 +1,7 @@
+# @epfml/discojs-node
+
+Decentralized & federated privacy-preserving ML training in TypeScript.
+
+This is the node-specific library of the Disco.js project,
+offering some helpers.
+You probably want to also use [`discojs`](../discojs).
diff --git a/discojs-web/README.md b/discojs-web/README.md
new file mode 100644
index 000000000..35a29482a
--- /dev/null
+++ b/discojs-web/README.md
@@ -0,0 +1,7 @@
+# @epfml/discojs-web
+
+Decentralized & federated privacy-preserving ML training in TypeScript.
+
+This is the web-specific library of the Disco.js project,
+offering some helpers.
+You probably want to also use [`discojs`](../discojs).
diff --git a/discojs/README.md b/discojs/README.md
new file mode 100644
index 000000000..7af371ce6
--- /dev/null
+++ b/discojs/README.md
@@ -0,0 +1,24 @@
+# @epfml/discojs
+
+Decentralized & federated privacy-preserving ML training in TypeScript.
+
+This is the core library of the Disco.js project.
+
+It is platform-agnostic, and has two companions library:
+ - [`discojs-node`](../discojs-node) for Node.js
+ - [`discojs-web`](../discojs-web) for web browsers
+
+The easiest way to start using it is through the `Disco` object.
+Create your own `Task` or load one from our `default_tasks`,
+setup the `Dataset` you want, and train with it.
+
+```ts
+import { Disco } from '@epfml/discojs'
+
+const url = ...; // url to a Disco.js server
+const dataset = ...;
+const task = ...;
+
+const disco = new Disco(task, { url })
+for await (const _ of disco.fit(dataset));
+```