As we move toward Agentic AI in ODC, we are currently forced to choose between two paths for vector storage:
External Vector DBs (Pinecone, Qdrant): Great performance, but introduces data residency/compliance hurdles for regulated industries.
Native Workarounds: Storing vectors as JSON text in ODC Entities.
I recently conducted a feasibility experiment to prove the "Native" path (storing vectors inside ODC to avoid external dependencies). You can see the architecture here:
https://medium.com/@michael.de.guzman/proving-vector-storage-retrieval-inside-outsystems-developer-cloud-a89d8fb88661
The Problem:
Currently, passing vector data between ODC and External Logic requires serializing arrays to JSON text. This introduces unnecessary overhead compared to passing native binary data. This creates a potential scalability bottleneck as dataset sizes grow.
The Solution:
Please introduce a Native Vector Data Type for ODC Entities and a built-in Vector_CosineSimilarity() server action (or Aggregate filter).
The Benefit:
This would allow ODC developers to build compliant, high-performance RAG and Agentic workflows entirely within the platform's trust boundary, without the massive overhead of JSON parsing.