Neon's PGrag Extension for Postgres RAG Pipelines

PGrag is an integration developed by Neon designed to facilitate end-to-end retrieval-augmented generation (RAG) pipelines directly within the Postgres environment [2]. This tool bridges vector databases and text embedding, allowing users to perform complex AI-driven tasks without leaving the database system [2]. To function, the extension relies on specific preloaded libraries that must be configured during the project setup or update process [1].

Key takeaways

• PGrag enables the construction of RAG pipelines directly inside Postgres, connecting vector databases and text embedding [2].
• The extension requires specific preloaded libraries, including rag_bge_small_en_v15 and rag_jina_reranker_v1_tiny_en [1].
• Users can exclude these libraries via API calls if they do not utilize the PGrag extension [1].
• The provided sources do not contain specific details regarding version 1.1.0.

Configuration and Library Requirements

Neon's implementation of PGrag necessitates the use of shared preload libraries, which are loaded into memory when the Postgres server starts [1]. Specifically, the extension utilizes the libraries rag_bge_small_en_v15 and rag_jina_reranker_v1_tiny_en [1]. These libraries are not enabled by default for all users; instead, they must be explicitly included in the preloaded_libraries object during a Create or Update project API call [1].

When configuring these libraries, users must specify all desired libraries in a single, comma-separated string. If a library is omitted from the API call, it will not be enabled [1]. Neon notes that the neon and pg_stat_statements libraries will remain enabled regardless of user configuration because they are utilized by the system-managed database [1].

Managing Extension Dependencies

Database administrators have the option to manage these libraries based on their project needs. If a project does not use the PGrag extension, the associated libraries can be excluded from the API call to configure the project [1]. It is important to note that Neon does not enable libraries for all extensions that have them, and available libraries may differ depending on the Postgres version being used [1].

Furthermore, Neon does not actively maintain third-party extension code, meaning bugs or issues should be reported to the upstream maintainers rather than Neon support [1]. While Neon supports the extension for installation and use, the underlying code is maintained externally.

Why it matters

Integrating RAG capabilities directly into Postgres via tools like