diff --git a/.changeset/per-cluster-centroid-dataset.md b/.changeset/per-cluster-centroid-dataset.md
new file mode 100644
index 0000000..c65b688
--- /dev/null
+++ b/.changeset/per-cluster-centroid-dataset.md
@@ -0,0 +1,6 @@
+---
+"@platforma-open/milaboratories.sequence-properties.workflow": patch
+"@platforma-open/milaboratories.sequence-properties.model": patch
+---
+
+Add support for per-cluster centroid datasets (clonotype-clustering). The model accepts a 2-axis `[sampleId, centroidId]` anchor, and the workflow processes the synthetic per-cluster consensus sequence in peptide mode (the upstream "Export consensus sequences as a dataset" feature is peptide-only, so the centroid sequence column is a peptide amino-acid column).
diff --git a/model/src/index.ts b/model/src/index.ts
index bc1eb47..0f9c56d 100644
--- a/model/src/index.ts
+++ b/model/src/index.ts
@@ -33,6 +33,16 @@ const inputAnchorSpecs = [
     axes: [{ name: "pl7.app/sampleId" }, { name: "pl7.app/vdj/scClonotypeKey" }],
     annotations: { "pl7.app/isAnchor": "true" },
   },
+  // Per-cluster centroid dataset (clonotype-clustering). A clonotype-shaped 2-axis
+  // dataset where the row entity is a cluster: axis[0] = sampleId (real per-sample
+  // abundance anchor), axis[1] = "pl7.app/clustering/centroidId" (values = real
+  // clusterId) carrying the synthetic per-cluster centroid sequence. The dedicated
+  // centroidId axis (not the producer's plain clusterId) keeps anchoring here from
+  // pulling its clusters table's centroid_*/reference_centroid_* columns.
+  {
+    axes: [{ name: "pl7.app/sampleId" }, { name: "pl7.app/clustering/centroidId" }],
+    annotations: { "pl7.app/isAnchor": "true" },
+  },
 ];
 
 export const platforma = BlockModelV3.create(blockDataModel)
diff --git a/workflow/src/main.tpl.tengo b/workflow/src/main.tpl.tengo
index 603fcf7..960105d 100644
--- a/workflow/src/main.tpl.tengo
+++ b/workflow/src/main.tpl.tengo
@@ -52,6 +52,17 @@ detectMode := func(axisSpec) {
 	if axisSpec.name == "pl7.app/vdj/scClonotypeKey" {
 		return "antibody_tcr_legacy_sc"
 	}
+	// Per-cluster centroid dataset (clonotype-clustering): a synthetic 2-axis
+	// [sampleId, centroidId] anchor carrying one consensus sequence per cluster. The
+	// "Export consensus sequences as a dataset" feature is peptide-only (the producer
+	// gates the export on peptide inputs), so the centroid sequence column is a peptide
+	// aminoacid column (pl7.app/sequence, pl7.app/feature: "peptide") and the dataset is
+	// processed in peptide mode — the peptideSequences matcher in prepare() picks it up.
+	// NOTE: if VDJ/single-cell centroid export is ever re-enabled upstream, this must
+	// branch on the centroid sequence column's feature instead of assuming peptide.
+	if axisSpec.name == "pl7.app/clustering/centroidId" {
+		return "peptide"
+	}
 	return ""
 }