主题
向量库里存放的是摄取后文档块的嵌入向量(有时也存放文档块本身)。
简单向量库
默认情况下,LlamaIndex 使用一个简单的内存向量库,非常适合快速试验。调用 vector_store.persist()(以及 SimpleVectorStore.from_persist_path(...))即可分别把它持久化到磁盘和从磁盘加载。
向量库选项与特性支持
LlamaIndex 支持 20 多种向量库。我们正在持续增加新的集成,并不断完善各家支持的特性覆盖。
| 向量库 | 类型 | 元数据过滤 | 混合检索 | 删除 | 存储文档 | 异步 |
|---|---|---|---|---|---|---|
| Alibaba Cloud OpenSearch | 云端 | ✓ | ✓ | ✓ | ✓ | |
| Apache Cassandra® | 自托管 / 云端 | ✓ | ✓ | ✓ | ||
| Astra DB | 云端 | ✓ | ✓ | ✓ | ||
| Azure AI Search | 云端 | ✓ | ✓ | ✓ | ✓ | |
| Azure CosmosDB Mongo vCore | 云端 | ✓ | ✓ | |||
| Azure CosmosDB NoSql | 云端 | ✓ | ✓ | |||
| BaiduVectorDB | 云端 | ✓ | ✓ | ✓ | ||
| ChatGPT Retrieval Plugin | 聚合器 | ✓ | ✓ | |||
| Chroma | 自托管 | ✓ | ✓ | ✓ | ||
| Couchbase | 自托管 / 云端 | ✓ | ✓ | ✓ | ✓ | |
| DashVector | 云端 | ✓ | ✓ | ✓ | ✓ | |
| Databricks | 云端 | ✓ | ✓ | ✓ | ||
| Deeplake | 自托管 / 云端 | ✓ | ✓ | ✓ | ||
| DocArray | 聚合器 | ✓ | ✓ | ✓ | ||
| DuckDB | 内存 / 自托管 | ✓ | ✓ | ✓ | ||
| DynamoDB | 云端 | ✓ | ||||
| Elasticsearch | 自托管 / 云端 | ✓ | ✓ | ✓ | ✓ | ✓ |
| FAISS | 内存 | |||||
| Google AlloyDB | 云端 | ✓ | ✓ | ✓ | ✓ | |
| Google Cloud SQL Postgres | 云端 | ✓ | ✓ | ✓ | ✓ | |
| Hnswlib | 内存 | |||||
| txtai | 内存 | |||||
| Jaguar | 自托管 / 云端 | ✓ | ✓ | ✓ | ✓ | |
| LanceDB | 云端 | ✓ | ✓ | ✓ | ||
| Lantern | 自托管 / 云端 | ✓ | ✓ | ✓ | ✓ | ✓ |
| MongoDB Atlas | 自托管 / 云端 | ✓ | ✓ | ✓ | ✓ | |
| MyScale | 云端 | ✓ | ✓ | ✓ | ✓ | |
| Milvus / Zilliz | 自托管 / 云端 | ✓ | ✓ | ✓ | ✓ | |
| Neo4jVector | 自托管 / 云端 | ✓ | ✓ | ✓ | ||
| OpenSearch | 自托管 / 云端 | ✓ | ✓ | ✓ | ✓ | ✓ |
| Pinecone | 云端 | ✓ | ✓ | ✓ | ✓ | |
| Postgres | 自托管 / 云端 | ✓ | ✓ | ✓ | ✓ | ✓ |
| pgvecto.rs | 自托管 / 云端 | ✓ | ✓ | ✓ | ✓ | |
| Qdrant | 自托管 / 云端 | ✓ | ✓ | ✓ | ✓ | ✓ |
| Redis | 自托管 / 云端 | ✓ | ✓ | ✓ | ||
| S3 | 云端 | ✓ | ✓ | ✓ | ✓* (借助 asyncio.to_thread) | |
| Simple | 内存 | ✓ | ✓ | |||
| SingleStore | 自托管 / 云端 | ✓ | ✓ | ✓ | ||
| Supabase | 自托管 / 云端 | ✓ | ✓ | ✓ | ||
| Tablestore | 云端 | ✓ | ✓ | ✓ | ✓ | |
| Tair | 云端 | ✓ | ✓ | ✓ | ||
| TiDB | 云端 | ✓ | ✓ | ✓ | ||
| TencentVectorDB | 云端 | ✓ | ✓ | ✓ | ✓ | |
| Timescale | ✓ | ✓ | ✓ | ✓ | ||
| Typesense | 自托管 / 云端 | ✓ | ✓ | ✓ | ||
| Upstash | 云端 | ✓ | ||||
| VectorX DB | 云端 | ✓ | ✓ | ✓ | ✓ | ✓ |
| Vearch | 自托管 | ✓ | ✓ | ✓ | ||
| Vespa | 自托管 / 云端 | ✓ | ✓ | ✓ | ✓ | |
| Vertex AI Vector Search | 云端 | ✓ | ✓ | ✓ | ||
| Weaviate | 自托管 / 云端 | ✓ | ✓ | ✓ | ✓ | |
| WordLift | 云端 | ✓ | ✓ | ✓ | ✓ | ✓ |
更多细节见向量库集成。
示例 Notebook
- Alibaba Cloud OpenSearch
- Astra DB
- Async Index Creation
- Azure AI Search
- Azure Cosmos DB Mongo vCore
- Azure Cosmos DB NoSql
- Baidu
- Caasandra
- Chromadb
- Couchbase
- Dash
- Databricks
- Deeplake
- DocArray HNSW
- DocArray in-Memory
- DuckDB
- Espilla
- Google AlloyDB for PostgreSQL
- Google Cloud SQL for PostgreSQL
- Jaguar
- LanceDB
- Lantern
- Milvus
- Milvus Async API
- Milvus Full-Text Search
- Milvus Hybrid Search
- MyScale
- ElasticSearch
- FAISS
- Hnswlib
- MongoDB Atlas
- Neo4j
- OpenSearch
- Pinecone
- Pinecone Hybrid Search
- PGvectoRS
- Postgres
- Redis
- Qdrant
- Qdrant Hybrid Search
- Rockset
- S3
- Simple
- Supabase
- Tablestore
- Tair
- TiDB
- Tencent
- Timesacle
- Upstash
- VectorX DB
- Vearch
- Vespa
- Vertex AI Vector Search
- Weaviate
- Weaviate Hybrid Search
- WordLift
- Zep