diff --git a/CHANGELOG.md b/CHANGELOG.md
index 109dc3cd..5f5ddbed 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -2,10 +2,15 @@
 
 ## [0.4.0a3] - ****-**-**
 
+**Features**
+
+- Document that the `zarr-vcf` dataset can be either a path or an in-memory zarr group.
+  (feature introduced in {pr}`966`, documented in {pr}`974`, {user}`hyanwong`)
+
 **Fixes**
 
 - Properly account for "N" as an unknown ancestral state, and ban "" from being
-  set as an ancestral state ({pr}`963`, {user}`hyanwong`))
+  set as an ancestral state ({pr}`963`, {user}`hyanwong`)
 
 ## [0.4.0a2] - 2024-09-06
 
diff --git a/tsinfer/formats.py b/tsinfer/formats.py
index 99d48064..c485d792 100644
--- a/tsinfer/formats.py
+++ b/tsinfer/formats.py
@@ -2308,8 +2308,10 @@ class VariantData(SampleData):
         the inference process will have ``inferred_ts.num_samples`` equal to double
         the number returned by ``VariantData.num_samples``.
 
-    :param str path: The path to the file containing the input dataset in VCF-Zarr
-        format.
+    :param Union(str, zarr.hierarchy.Group) path_or_zarr: The input dataset in
+        `VCF Zarr <https://github.com/sgkit-dev/vcf-zarr-spec>`_ format.
+        This can either a path to the Zarr dataset saved on disk, or the
+        Zarr object itself.
     :param Union(array, str) ancestral_state: A numpy array of strings specifying
         the ancestral states (alleles) used in inference. This must be the same length
         as the number of unmasked sites in the dataset. Alternatively, a single string