Vector Store
A system for storing and retrieving vector data, constituting one form of implementation of a vector database.
Timeline (last 90 days)
Under float32 storage, each 1024-dimensional vector occupies 4KB, so an index of a million-scale documents requires at least 4GB of storage.
For continual learning that does not involve weight updates (long context, memory files, vector stores), the risk is lower; text memories are readable, and developers can read and delete content that describes evading the monitor.
Dedicating the vector store to semantic recall while maintaining a separate relational table for success rate statistics linked via work order ID offers clear separation of responsibilities and more efficient aggregation queries, at the cost of needing to keep writes to both stores synchronized when work orders are archived
When multiple agents share the same vector store, there is no natural isolation mechanism, and any agent may retrieve private information belonging to other agents.
When multiple agents share the same vector store, the memories, conversation histories, and knowledge bases of all agents are mixed together in a single data pool, where any agent may retrieve the private information of other agents, with no isolation mechanism in place.
When multiple agents share the same vector store without an isolation mechanism, any agent may retrieve private information from other agents, and once one agent is compromised, the attack surface covers the entire shared store.
All Facts (6)
Under float32 storage, each 1024-dimensional vector occupies 4KB, so an index of a million-scale documents requires at least 4GB of storage.
50%UnverifiedFor continual learning that does not involve weight updates (long context, memory files, vector stores), the risk is lower; text memories are readable, and developers can read and delete content that describes evading the monitor.
50%UnverifiedDedicating the vector store to semantic recall while maintaining a separate relational table for success rate statistics linked via work order ID offers clear separation of responsibilities and more efficient aggregation queries, at the cost of needing to keep writes to both stores synchronized when work orders are archived
50%UnverifiedWhen multiple agents share the same vector store, there is no natural isolation mechanism, and any agent may retrieve private information belonging to other agents.
50%UnverifiedWhen multiple agents share the same vector store, the memories, conversation histories, and knowledge bases of all agents are mixed together in a single data pool, where any agent may retrieve the private information of other agents, with no isolation mechanism in place.
50%UnverifiedWhen multiple agents share the same vector store without an isolation mechanism, any agent may retrieve private information from other agents, and once one agent is compromised, the attack surface covers the entire shared store.
50%