diff --git a/mypyc/doc/index.rst b/mypyc/doc/index.rst
index 8be1086bdd70..aacf275de988 100644
--- a/mypyc/doc/index.rst
+++ b/mypyc/doc/index.rst
@@ -33,6 +33,7 @@ generate fast code.
 
    librt
    librt_base64
+   librt_random
    librt_strings
    librt_time
    librt_vecs
diff --git a/mypyc/doc/librt.rst b/mypyc/doc/librt.rst
index a9492e1d6126..23206f8cfe80 100644
--- a/mypyc/doc/librt.rst
+++ b/mypyc/doc/librt.rst
@@ -26,6 +26,8 @@ Follow submodule links in the table to a detailed description of each submodule.
      - Description
    * - :doc:`librt.base64 <librt_base64>`
      - Fast Base64 encoding and decoding
+   * - :doc:`librt.random <librt_random>`
+     - Pseudorandom number generation
    * - :doc:`librt.strings <librt_strings>`
      - String and bytes utilities
    * - :doc:`librt.time <librt_time>`
diff --git a/mypyc/doc/librt_random.rst b/mypyc/doc/librt_random.rst
new file mode 100644
index 000000000000..d5543661ce98
--- /dev/null
+++ b/mypyc/doc/librt_random.rst
@@ -0,0 +1,97 @@
+.. _librt-random:
+
+librt.random
+============
+
+The ``librt.random`` module is part of the ``librt`` package on PyPI, and it provides
+pseudorandom number generation utilities. It can be used as a significantly faster
+alternative to the stdlib :mod:`random` module in compiled code. It can also be faster
+than stdlib ``random`` in interpreted code, depending on use case.
+
+The module uses the `ChaCha8 <https://cr.yp.to/chacha.html>`__ algorithm with forward
+secrecy. It is **not** suitable for cryptographic use, but it provides high-quality,
+statistically uniform output.
+
+Functions
+---------
+
+The module provides module-level functions that use thread-local state, so they are
+safe to call concurrently from multiple threads without external locking, and they
+scale well even if used from multiple threads:
+
+.. function:: random() -> float
+
+   Return a random floating-point number in the range [0.0, 1.0).
+
+.. function:: randint(a: i64, b: i64) -> i64
+
+   Return a random integer *n* such that *a* <= *n* <= *b*.
+
+.. function:: randrange(stop: i64, /) -> i64
+              randrange(start: i64, stop: i64, /) -> i64
+
+   Return a random integer from the range. With one argument, the range is [0, *stop*).
+   With two arguments, the range is [*start*, *stop*).
+
+.. function:: seed(n: i64, /) -> None
+
+   Seed the thread-local random number generator. This only affects module-level
+   functions called from the current thread.
+
+Random class
+------------
+
+.. class:: Random(seed: i64 | None = None)
+
+   A pseudorandom number generator instance with its own independent state. Use this
+   when you need reproducible sequences or want to avoid interference with the
+   thread-local state used by the module-level functions.
+
+   If *seed* is ``None``, the generator is seeded from OS entropy
+   (via :func:`os.urandom`).
+
+   It's not safe to use the same ``Random`` instance concurrently from multiple
+   threads without synchronization on free-threaded Python builds.
+
+   .. method:: random() -> float
+
+      Return a random floating-point number in the range [0.0, 1.0).
+
+   .. method:: randint(a: i64, b: i64) -> i64
+
+      Return a random integer *n* such that *a* <= *n* <= *b*.
+
+   .. method:: randrange(stop: i64, /) -> i64
+               randrange(start: i64, stop: i64, /) -> i64
+
+      Return a random integer from the range. With one argument, the range is [0, *stop*).
+      With two arguments, the range is [*start*, *stop*).
+
+   .. method:: seed(n: i64, /) -> None
+
+      Reseed the generator.
+
+Example
+-------
+
+Using module-level functions::
+
+    from librt.random import randint, seed
+
+    def roll_dice() -> i64:
+        return randint(1, 6)
+
+Using a ``Random`` instance for reproducible sequences::
+
+    from librt.random import Random
+
+    def generate_data() -> list[i64]:
+        rng = Random(42)
+        return [rng.randint(0, 100) for _ in range(10)]
+
+Backward compatibility
+----------------------
+
+New versions of this module are not guaranteed to generate the same results when
+using the same seed. A specific seed only produces predictable random numbers on a
+specific version of ``librt``. In the future we might provide stronger guarantees.