diff --git a/src/matrix/storage/idb/stores/RoomFragmentStore.js b/src/matrix/storage/idb/stores/RoomFragmentStore.js
new file mode 100644
index 00000000..86bd243f
--- /dev/null
+++ b/src/matrix/storage/idb/stores/RoomFragmentStore.js
@@ -0,0 +1,46 @@
+export default class RoomFragmentStore {
+    constructor(store) {
+        this._store = store;
+    }
+
+    _allRange(roomId) {
+        return IDBKeyRange.bound(
+            [roomId, Number.MIN_SAFE_INTEGER],
+            [roomId, Number.MAX_SAFE_INTEGER]
+        );
+    }
+
+    all(roomId) {
+        return this._store.selectAll(this._allRange(roomId));
+    }
+
+    /** Returns the fragment without a nextToken and without nextId,
+    if any, with the largest id if there are multiple (which should not happen) */
+    liveFragment(roomId) {
+        // why do we need this?
+        // Ok, take the case where you've got a /context fragment and a /sync fragment
+        // They are not connected. So, upon loading the persister, which one do we take? We can't sort them ...
+        // we assume that the one without a nextToken and without a nextId is a live one
+        // there should really be only one like this
+
+        // reverse because assuming live fragment has bigger id than non-live ones
+        return this._store.findReverse(this._allRange(roomId), fragment => {
+            return typeof fragment.nextId !== "number" && typeof fragment.nextToken !== "string";
+        });
+    }
+
+    // should generate an id an return it?
+    // depends if we want to do anything smart with fragment ids,
+    // like give them meaning depending on range. not for now probably ...
+    add(fragment) {
+        return this._store.add(fragment);
+    }
+
+    update(fragment) {
+        return this._store.put(fragment);
+    }
+
+    get(roomId, fragmentId) {
+        return this._store.get([roomId, fragmentId]);
+    }
+}