diff --git a/nexus/mnemosyne/linker.py b/nexus/mnemosyne/linker.py
index f1f68d62..3b4309c8 100644
--- a/nexus/mnemosyne/linker.py
+++ b/nexus/mnemosyne/linker.py
@@ -2,31 +2,63 @@
 
 Computes semantic similarity between archive entries and creates
 bidirectional links, forming the holographic graph structure.
+
+Supports pluggable embedding backends for true semantic search.
+Falls back to Jaccard token similarity when no backend is available.
 """
 
 from __future__ import annotations
 
-from typing import Optional
+from typing import Optional, TYPE_CHECKING
+
 from nexus.mnemosyne.entry import ArchiveEntry
 
+if TYPE_CHECKING:
+    from nexus.mnemosyne.embeddings import EmbeddingBackend
+
 
 class HolographicLinker:
     """Links archive entries via semantic similarity.
 
-    Phase 1 uses simple keyword overlap as the similarity metric.
-    Phase 2 will integrate ChromaDB embeddings from MemPalace.
+    With an embedding backend: cosine similarity on vectors.
+    Without: Jaccard similarity on token sets (legacy fallback).
     """
 
-    def __init__(self, similarity_threshold: float = 0.15):
+    def __init__(
+        self,
+        similarity_threshold: float = 0.15,
+        embedding_backend: Optional["EmbeddingBackend"] = None,
+    ):
         self.threshold = similarity_threshold
+        self._backend = embedding_backend
+        self._embed_cache: dict[str, list[float]] = {}
+
+    @property
+    def using_embeddings(self) -> bool:
+        return self._backend is not None
+
+    def _get_embedding(self, entry: ArchiveEntry) -> list[float]:
+        """Get or compute cached embedding for an entry."""
+        if entry.id in self._embed_cache:
+            return self._embed_cache[entry.id]
+        text = f"{entry.title} {entry.content}"
+        vec = self._backend.embed(text) if self._backend else []
+        if vec:
+            self._embed_cache[entry.id] = vec
+        return vec
 
     def compute_similarity(self, a: ArchiveEntry, b: ArchiveEntry) -> float:
         """Compute similarity score between two entries.
 
-        Returns float in [0, 1]. Phase 1: Jaccard similarity on
-        combined title+content tokens. Phase 2: cosine similarity
-        on ChromaDB embeddings.
+        Returns float in [0, 1]. Uses embedding cosine similarity if
+        a backend is configured, otherwise falls back to Jaccard.
         """
+        if self._backend:
+            vec_a = self._get_embedding(a)
+            vec_b = self._get_embedding(b)
+            if vec_a and vec_b:
+                return self._backend.similarity(vec_a, vec_b)
+        # Fallback: Jaccard on tokens
         tokens_a = self._tokenize(f"{a.title} {a.content}")
         tokens_b = self._tokenize(f"{b.title} {b.content}")
         if not tokens_a or not tokens_b:
@@ -35,11 +67,10 @@ class HolographicLinker:
         union = tokens_a | tokens_b
         return len(intersection) / len(union)
 
-    def find_links(self, entry: ArchiveEntry, candidates: list[ArchiveEntry]) -> list[tuple[str, float]]:
-        """Find entries worth linking to.
-
-        Returns list of (entry_id, similarity_score) tuples above threshold.
-        """
+    def find_links(
+        self, entry: ArchiveEntry, candidates: list[ArchiveEntry]
+    ) -> list[tuple[str, float]]:
+        """Find entries worth linking to. Returns (entry_id, score) tuples."""
         results = []
         for candidate in candidates:
             if candidate.id == entry.id:
@@ -58,16 +89,18 @@ class HolographicLinker:
             if eid not in entry.links:
                 entry.links.append(eid)
                 new_links += 1
-            # Bidirectional
             for c in candidates:
                 if c.id == eid and entry.id not in c.links:
                     c.links.append(entry.id)
         return new_links
 
+    def clear_cache(self):
+        """Clear embedding cache (call after bulk entry changes)."""
+        self._embed_cache.clear()
+
     @staticmethod
     def _tokenize(text: str) -> set[str]:
         """Simple whitespace + punctuation tokenizer."""
         import re
         tokens = set(re.findall(r"\w+", text.lower()))
-        # Remove very short tokens
         return {t for t in tokens if len(t) > 2}