Vector database
Enterprise h2oGPTe uses a vector database to store document chunks for future retrieval. The following section describes the currently supported options.
Embedded Vex vector database​
This option is the default embedded vector database that ships out of the box with the software.
Vex == uvicorn + hnswlib + sqlite
Hnswlib (fast approximate nearest neighbor search) is projected to encounter performance degradation at roughly 10M-50M entries.
A separate (hnswlib+sqlite) instance is created per Collection, so those limits apply at the Collection level.
External vector databases​
The following external vector databases are currently experimentally supported:
- ElasticSearch
- Milvus
- Qdrant
- Redis
Vector database == uvicorn + [ ElasticSearch / Milvus / Qdrant / Redis ]
Please contact us if your preferred vector database is not listed above (these can be added based on customer interest).
- Submit and view feedback for this page
- Send feedback about Enterprise h2oGPTe to cloud-feedback@h2o.ai