diff --git a/ADRs/003_rocksdb.md b/ADRs/003_rocksdb.md
new file mode 100644
index 000000000..c274217a6
--- /dev/null
+++ b/ADRs/003_rocksdb.md
@@ -0,0 +1,38 @@
+# RocksDB ADR
+
+**Author**
+Marko Budiselic (github.com/gitbuda)
+
+**Status**
+ACCEPTED
+
+**Date**
+January 23, 2024
+
+**Problem**
+
+Interacting with data (reads and writes) on disk in a concurrent, safe, and
+fast way is a challenging task. Implementing all low-level primitives to
+interact with various disk hardware efficiently consumes significant
+engineering people. Whenever Memgraph has to store data on disk (or any
+other colder than RAM storage system), the problem is how to do that in the
+least amount of development time while satisfying all functional requirements
+(often performance).
+
+**Criteria**
+
+- working efficiently in a highly concurrent environment
+- easy integration with Memgraph's C++ codebase
+- providing low-level key-value API
+- heavily tested in production environments
+- providing abstractions for the storage hardware (even for cloud-based
+  storages like S3)
+
+**Decision**
+
+There are a few robust key-value stores, but finding one that is
+production-ready and compatible with Memgraph's C++ codebase is challenging.
+**We select [RocksDB](https://github.com/facebook/rocksdb)** because it
+delivers robust API to manage data on disk; it's battle-tested in many
+production environments (many databases systems are embedding RocksDB), and
+it's the most compatible one.