Initial commit: standalone vector memory server

REST API for per-agent semantic memory retrieval.

- FastAPI server with /api/{agent}/save and /api/{agent}/query
- ChromaDB for persistent vector storage
- all-MiniLM-L6-v2 via sentence-transformers for embeddings
- Per-agent collections for clean separation
- Config through env vars or config.py
- .venv ready with all dependencies

embedder.py

@@ -0,0 +1,29 @@
+"""Embedding model wrapper — singleton pattern.
+
+Uses sentence-transformers with all-MiniLM-L6-v2 (384-dim, ~80MB).
+Loaded once, reused across all requests.
+"""
+
+from sentence_transformers import SentenceTransformer
+from config import Config
+
+
+class Embedder:
+    """Thread-safe singleton wrapper for the embedding model."""
+
+    _instance = None
+
+    @classmethod
+    def get_instance(cls) -> "Embedder":
+        if cls._instance is None:
+            cls._instance = cls()
+        return cls._instance
+
+    def __init__(self) -> None:
+        self.model = SentenceTransformer(Config.EMBEDDING_MODEL)
+
+    def embed(self, text: str) -> list[float]:
+        return self.model.encode(text).tolist()
+
+    def embed_batch(self, texts: list[str]) -> list[list[float]]:
+        return self.model.encode(texts).tolist()