• When initializing the Qdrant client, a global Client can be initialized to handle CRUD operations for collections.
• Filter vectors using metadata.
• Delete relevant collections by performing a fuzzy search on collection names (in conjunction with Redis).
• Check if a collection exists by performing a fuzzy search on collection names (in conjunction with Redis) and matching the full name.

For the above requirements, consider the distinction and unification between local storage and self-hosted cloud deployment.

• When initializing the Qdrant client, a global Client can be initialized to handle CRUD operations for collections.
• Filter vectors using metadata.
• Delete relevant collections by performing a fuzzy search on collection names (in conjunction with Redis).
• Check if a collection exists by performing a fuzzy search on collection names (in conjunction with Redis) and matching the full name.

For the above requirements, consider the distinction and unification between local storage and self-hosted cloud deployment.

all doable

thanks @Wendong-Fan , and I will take the task and develop

• When initializing the Qdrant client, a global Client can be initialized to handle CRUD operations for collections.
• Filter vectors using metadata.
• Delete relevant collections by performing a fuzzy search on collection names (in conjunction with Redis).
• Check if a collection exists by performing a fuzzy search on collection names (in conjunction with Redis) and matching the full name.

For the above requirements, consider the distinction and unification between local storage and self-hosted cloud deployment.

I have completed the development and have a few clarifications regarding the requirements:

• Collections cannot be filtered by metadata, and it's not recommended to integrate Redis for this purpose. Qdrant tools should maintain loose coupling.
• Collections should not be deleted based solely on fuzzy search for collection types, as this approach is too risky. It’s necessary to ensure that the full collection_name is used to delete a collection.
• If there are many collections with similar names, I think the vectors should be placed in the same collection, and classification should be handled using payload.
• Self-hosted mode actually doesn't support concurrent searches. In the original design, this was addressed by managing through an in-memory list, which I believe is sufficient for local testing. In server mode, this issue doesn't need to be considered. Therefore, I believe the functionality between local storage and self-hosted Qdrant has been unified.