feat: Adds Vector Search With Qdrant

README.md

@@ -28,7 +28,7 @@ Agent conversational systems rely on limited context windows — once the Token
 
 - **Semantic Retrieval & Conflict Detection**: Unified cross-type vector semantic retrieval; `MemUpdateChecker` uses LLM to analyze semantic conflicts and intelligently decide ADD/DELETE strategies; LLM outputs UPDATE/DELETE directives validated via semantic checks before execution, ensuring memory consistency and controllable operations.
 
-- **Full-Stack Storage Backend System**: Coverage across five storage categories — KV (InMemoryKV/ShelveStore/DbBasedKV/Redis), Vector (ChromaDB/Milvus/Elasticsearch/GaussVector), Relational (SQLite/PostgreSQL/MySQL/GaussDB), Message (SqlMessageStore), and Graph (Milvus GraphStore) — adapting from local single-node to cloud cluster scenarios.
+- **Full-Stack Storage Backend System**: Coverage across five storage categories — KV (InMemoryKV/ShelveStore/DbBasedKV/Redis), Vector (ChromaDB/Milvus/Elasticsearch/GaussVector/Qdrant), Relational (SQLite/PostgreSQL/MySQL/GaussDB), Message (SqlMessageStore), and Graph (Milvus GraphStore) — adapting from local single-node to cloud cluster scenarios.
 
 - **Data Migration Framework**: Supports versioned schema migration for KV/vector/SQL/message/index stores and cross-BaseMemoryIndex batch data migration, with an operation registry for custom migration extensions.
 
@@ -80,6 +80,9 @@ pip install JiuwenMemory[redis]
 # ChromaDB vector store
 pip install JiuwenMemory[chromadb]
 
+# Qdrant vector store
+pip install JiuwenMemory[qdrant]
+
 # File-system memory backend (sqlite-vec + watchdog + jieba)
 # Required for INDEX_BACKEND=file; missing deps silently degrade:
 pip install JiuwenMemory[file-index]
@@ -330,7 +333,7 @@ Graph Memory is an independent knowledge graph memory module. It turns input con
 
 ### **Flexible Storage Backends and Data Migration**
 
-- **Full-Stack Storage Backends**: Coverage across five storage categories — KV (InMemoryKV/ShelveStore/DbBasedKV/Redis), Vector (ChromaDB/Milvus/Elasticsearch/GaussVector), Relational (SQLite/PostgreSQL/MySQL/GaussDB), Message (SqlMessageStore), and Graph (Milvus GraphStore) — adapting from local single-node to cloud cluster scenarios.
+- **Full-Stack Storage Backends**: Coverage across five storage categories — KV (InMemoryKV/ShelveStore/DbBasedKV/Redis), Vector (ChromaDB/Milvus/Elasticsearch/GaussVector/Qdrant), Relational (SQLite/PostgreSQL/MySQL/GaussDB), Message (SqlMessageStore), and Graph (Milvus GraphStore) — adapting from local single-node to cloud cluster scenarios.
 - **Versioned Migration**: A complete migration framework supporting SQL schema changes, vector field renaming, KV data updates, message data transformation, and index field operations.
 - **Cross-index Migration**: Supports batch migration of memory data between different BaseMemoryIndex instances for smooth storage engine switching, with an operation registry for custom migration extensions.
 

README.zh.md

@@ -28,7 +28,7 @@ Agent对话系统依赖有限的上下文窗口，一旦超出 Token 限制或
 
 - **语义检索与冲突检测**：跨记忆类型统一向量语义检索；`MemUpdateChecker` 通过 LLM 分析语义冲突智能决定 ADD/DELETE 策略，LLM 输出 UPDATE/DELETE 指令经语义校验后执行，确保记忆一致与操作可控。
 
-- **全栈存储后端体系**：覆盖 KV（InMemoryKV/ShelveStore/DbBasedKV/Redis）、向量（ChromaDB/Milvus/Elasticsearch/GaussVector）、关系型（SQLite/PostgreSQL/MySQL/GaussDB）、消息（SqlMessageStore）、图（Milvus GraphStore）五大存储类别，适配本地单机到云端集群全场景。
+- **全栈存储后端体系**：覆盖 KV（InMemoryKV/ShelveStore/DbBasedKV/Redis）、向量（ChromaDB/Milvus/Elasticsearch/GaussVector/Qdrant）、关系型（SQLite/PostgreSQL/MySQL/GaussDB）、消息（SqlMessageStore）、图（Milvus GraphStore）五大存储类别，适配本地单机到云端集群全场景。
 
 - **数据迁移框架**：支持 KV/向量/SQL/消息/索引的版本化 schema 迁移与跨 BaseMemoryIndex 实例批量数据迁移，操作注册表支持自定义迁移扩展。
 
@@ -80,6 +80,9 @@ pip install JiuwenMemory[redis]
 # ChromaDB 向量存储
 pip install JiuwenMemory[chromadb]
 
+# Qdrant 向量存储
+pip install JiuwenMemory[qdrant]
+
 # 文件系统记忆后端（sqlite-vec + watchdog + jieba）
 # 用 INDEX_BACKEND=file 时需要，缺失则静默降级
 pip install JiuwenMemory[file-index]
@@ -322,7 +325,7 @@ Graph Memory 是独立的知识图谱记忆模块，可将输入内容沉淀为
 
 ### **灵活的存储后端与数据迁移**
 
-- **全栈存储后端**：覆盖 KV（InMemoryKV/ShelveStore/DbBasedKV/Redis）、向量（ChromaDB/Milvus/Elasticsearch/GaussVector）、关系型（SQLite/PostgreSQL/MySQL/GaussDB）、消息（SqlMessageStore）、图（Milvus GraphStore）五大存储类别，适配本地单机到云端集群全场景。
+- **全栈存储后端**：覆盖 KV（InMemoryKV/ShelveStore/DbBasedKV/Redis）、向量（ChromaDB/Milvus/Elasticsearch/GaussVector/Qdrant）、关系型（SQLite/PostgreSQL/MySQL/GaussDB）、消息（SqlMessageStore）、图（Milvus GraphStore）五大存储类别，适配本地单机到云端集群全场景。
 - **版本化迁移**：提供完整的迁移框架，支持 SQL schema 变更、向量字段重命名、KV 数据更新、消息数据转换和索引字段操作等多种迁移类型。
 - **跨索引迁移**：支持在不同 BaseMemoryIndex 实例之间批量迁移记忆数据，操作注册表支持自定义迁移扩展，实现存储引擎的平滑切换。
 

docs/en/API Docs/memory_server.md

@@ -105,7 +105,7 @@ If neither `.env` file is found, the service automatically creates the `~/.jiuwe
 | `DB_STORE_TYPE` | `default` | DB Store type. Supported values: `default` / `gauss`. |
 | `INDEX_BACKEND` | `simple` | Memory index backend. Supported values: `simple` (vector, KV+Vector) / `file` (markdown+SQLite). When `file`, long-term memories are persisted to markdown files; see `FILE_MEMORY_DATA_DIR`. |
 | `FILE_MEMORY_DATA_DIR` | `~/.jiuwenmemory/file_memory_data` | `FileMemoryIndex` data root directory. Only effective when `INDEX_BACKEND=file`. |
-| `VECTOR_STORE_TYPE` | `chroma` | Vector Store type. Supported values: `chroma` / `milvus` / `elasticsearch` / `gauss`. Under `INDEX_BACKEND=file`, the vector store is only used for middle-term memory / dreaming and may be left unconfigured. |
+| `VECTOR_STORE_TYPE` | `chroma` | Vector Store type. Supported values: `chroma` / `milvus` / `elasticsearch` / `gauss` / `qdrant`. Under `INDEX_BACKEND=file`, the vector store is only used for middle-term memory / dreaming and may be left unconfigured. |
 | `VECTOR_CHROMA_PERSIST_DIR` | `MEMORY_DATA_DIR` | Chroma persistence directory. |
 | `VECTOR_MILVUS_URI` | empty string | Milvus service URI. |
 | `VECTOR_MILVUS_TOKEN` | empty string | Milvus token; optional. |
@@ -124,6 +124,11 @@ If neither `.env` file is found, the service automatically creates the `~/.jiuwe
 | `VECTOR_GAUSS_DATABASE` | `postgres` | Gauss vector store database name. |
 | `VECTOR_GAUSS_USER` | `postgres` | Gauss vector store user. |
 | `VECTOR_GAUSS_PASSWORD` | empty string | Gauss vector store password. |
+| `VECTOR_QDRANT_URL` | `http://localhost:6333` | Qdrant HTTP endpoint. |
+| `VECTOR_QDRANT_API_KEY` | empty string | Qdrant API key; optional for unsecured local deployments. |
+| `VECTOR_QDRANT_COLLECTION_PREFIX` | `agent_vector` | Prefix used to isolate JiuwenMemory collections. |
+| `VECTOR_QDRANT_PREFER_GRPC` | `false` | Prefer Qdrant's gRPC transport when set to `true`. |
+| `VECTOR_QDRANT_TIMEOUT` | client default | Request timeout in seconds; leave empty to use the client default. |
 
 #### `INDEX_BACKEND=file` notes
 

docs/zh/API文档/memory_server.md

@@ -105,7 +105,7 @@ IP=127.0.0.1 PORT=8000 python -m jiuwen_memory.server.memory_server
 | `DB_STORE_TYPE` | `default` | DB Store 类型，支持 `default` / `gauss`。 |
 | `INDEX_BACKEND` | `simple` | 记忆索引后端，支持 `simple`（向量，KV+Vector）/ `file`（markdown+SQLite）。`file` 时长期记忆落 markdown 文件，配合 `FILE_MEMORY_DATA_DIR`。 |
 | `FILE_MEMORY_DATA_DIR` | `~/.jiuwenmemory/file_memory_data` | FileMemoryIndex 数据根目录，仅 `INDEX_BACKEND=file` 时生效。 |
-| `VECTOR_STORE_TYPE` | `chroma` | Vector Store 类型，支持 `chroma` / `milvus` / `elasticsearch` / `gauss`。`INDEX_BACKEND=file` 时仅用于中间记忆 / dreaming，可不配。 |
+| `VECTOR_STORE_TYPE` | `chroma` | Vector Store 类型，支持 `chroma` / `milvus` / `elasticsearch` / `gauss` / `qdrant`。`INDEX_BACKEND=file` 时仅用于中间记忆 / dreaming，可不配。 |
 | `VECTOR_CHROMA_PERSIST_DIR` | `MEMORY_DATA_DIR` | Chroma 向量库持久化目录。 |
 | `VECTOR_MILVUS_URI` | 空字符串 | Milvus 服务地址。 |
 | `VECTOR_MILVUS_TOKEN` | 空字符串 | Milvus Token，可为空。 |
@@ -124,6 +124,11 @@ IP=127.0.0.1 PORT=8000 python -m jiuwen_memory.server.memory_server
 | `VECTOR_GAUSS_DATABASE` | `postgres` | Gauss 向量库数据库名。 |
 | `VECTOR_GAUSS_USER` | `postgres` | Gauss 向量库用户名。 |
 | `VECTOR_GAUSS_PASSWORD` | 空字符串 | Gauss 向量库密码。 |
+| `VECTOR_QDRANT_URL` | `http://localhost:6333` | Qdrant HTTP 服务地址。 |
+| `VECTOR_QDRANT_API_KEY` | 空字符串 | Qdrant API Key；本地无鉴权部署可留空。 |
+| `VECTOR_QDRANT_COLLECTION_PREFIX` | `agent_vector` | JiuwenMemory collection 隔离前缀。 |
+| `VECTOR_QDRANT_PREFER_GRPC` | `false` | 设为 `true` 时优先使用 Qdrant gRPC 传输。 |
+| `VECTOR_QDRANT_TIMEOUT` | 客户端默认值 | 请求超时秒数；留空使用客户端默认值。 |
 
 #### `INDEX_BACKEND=file` 说明
 

jiuwen_memory/foundation/store/__init__.py

@@ -32,7 +32,7 @@
 
 # Built-in backends. Closed to extension by design — use register_vector_store
 # or the entry_points mechanism for 3rd-party backends.
-_BUILTIN_VECTOR_STORE_NAMES = frozenset({"chroma", "milvus", "gaussvector", "elasticsearch"})
+_BUILTIN_VECTOR_STORE_NAMES = frozenset({"chroma", "milvus", "gaussvector", "elasticsearch", "qdrant"})
 
 # Explicit in-process registrations (register_vector_store).
 # Maps backend name -> factory callable (typically a class).
@@ -48,7 +48,7 @@ def register_vector_store(
     Use this in application init code when shipping a plugin via
     entry_points is not practical (e.g., private in-repo backend).
 
-    Built-in names (chroma, milvus, gaussvector) cannot be overridden — a
+    Built-in names (chroma, milvus, gaussvector, elasticsearch, qdrant) cannot be overridden — a
     register call with a built-in name is kept in the registry but the
     built-in still wins in ``create_vector_store()`` resolution.
 
@@ -78,6 +78,9 @@ def _resolve_builtin(store_type: str, kwargs: dict) -> "BaseVectorStore | None":
     if store_type == "elasticsearch":
         from jiuwen_memory.foundation.store.vector.es_vector_store import ElasticsearchVectorStore
         return ElasticsearchVectorStore(**kwargs)
+    if store_type == "qdrant":
+        from jiuwen_memory.foundation.store.vector.qdrant_vector_store import QdrantVectorStore
+        return QdrantVectorStore(**kwargs)
     return None
 
 
@@ -122,7 +125,7 @@ def create_vector_store(store_type: str, **kwargs) -> "BaseVectorStore | None":
     """Factory for vector-store backends.
 
     Resolution order:
-      1. Built-in (chroma, milvus, gaussvector) — always wins, closed set.
+      1. Built-in (chroma, milvus, gaussvector, elasticsearch, qdrant) — always wins, closed set.
       2. Explicit registrations via ``register_vector_store()``.
       3. Entry_points in group ``openjiuwen.vector_stores``.