初始化提交

backend/app/integrations/chromadb/memory_store.py

@@ -0,0 +1,93 @@
+"""纯内存向量存储后端，供测试用。
+
+不持久化、不依赖外部服务，每次进程重启清空。
+"""
+
+from __future__ import annotations
+
+from typing import Any
+
+
+class MemoryVectorStore:
+    """纯内存向量存储（测试用）。"""
+
+    def __init__(self) -> None:
+        self._data: dict[str, dict[str, dict[str, Any]]] = {}
+        """{character_id: {memory_id: {"text": ..., "embedding": [...], "metadata": {...}}}}"""
+
+    # ── 增 ────────────────────────────────────────────────────────────────
+
+    def add_memory(
+        self,
+        character_id: str,
+        memory_id: str,
+        text: str,
+        embedding: list[float],
+        *,
+        metadata: dict[str, Any] | None = None,
+    ) -> None:
+        char_store = self._data.setdefault(character_id, {})
+        if memory_id not in char_store:
+            char_store[memory_id] = {}
+        char_store[memory_id] = {
+            "text": text,
+            "embedding": embedding,
+            "metadata": metadata or {},
+        }
+
+    def update_memory(
+        self,
+        character_id: str,
+        memory_id: str,
+        text: str,
+        embedding: list[float],
+        *,
+        metadata: dict[str, Any] | None = None,
+    ) -> None:
+        self.add_memory(character_id, memory_id, text, embedding, metadata=metadata)
+
+    def delete_memory(self, character_id: str, memory_id: str) -> None:
+        char_store = self._data.get(character_id)
+        if char_store:
+            char_store.pop(memory_id, None)
+
+    # ── 查 ────────────────────────────────────────────────────────────────
+
+    def search(
+        self,
+        character_id: str,
+        query: list[float],
+        *,
+        limit: int = 5,
+    ) -> list[dict[str, Any]]:
+        """余弦相似度暴力搜索（O(n)，仅测试用）。"""
+        char_store = self._data.get(character_id)
+        if not char_store:
+            return []
+
+        scored: list[tuple[float, str, dict[str, Any]]] = []
+        for mid, entry in char_store.items():
+            emb = entry.get("embedding", [])
+            if not emb:
+                continue
+            score = _cosine_similarity(query, emb)
+            scored.append((score, mid, entry.get("metadata", {})))
+
+        scored.sort(key=lambda x: x[0], reverse=True)
+        return [
+            {"id": mid, "score": score, "metadata": meta}
+            for score, mid, meta in scored[:limit]
+        ]
+
+    def count(self, character_id: str) -> int:
+        char_store = self._data.get(character_id)
+        return len(char_store) if char_store else 0
+
+
+def _cosine_similarity(a: list[float], b: list[float]) -> float:
+    dot = sum(x * y for x, y in zip(a, b, strict=False))
+    norm_a = sum(x * x for x in a) ** 0.5
+    norm_b = sum(x * x for x in b) ** 0.5
+    if norm_a == 0 or norm_b == 0:
+        return 0.0
+    return dot / (norm_a * norm_b)