Agentic AI is a platform for building and deploying AI agents that run directly in your Postgres database. It provides an agent UI, model serving, and managed orchestration for teams that want to build RAG applications without managing AI infrastructure separately.
It enables you to store, index, and search complex data like text or images by transforming them into mathematical coordinates, powering Retrieval-Augmented Generation (RAG) and semantic search applications directly inside your Postgres database.
AIDB
AIDB acts as the orchestrator of your AI workflows. It automates the complex backend tasks required to make your data AI-ready:
In-Database LLM integration: Connect directly to OpenAI, Azure, Google Cloud Storage, AWS S3, or local models using SQL.
AI data preparation: Automatically transform table data into vector embeddings.
Semantic management: No more glue code, you can manage your RAG workflows with standard SQL commands.
For more information, see AIDB docs.
pgvector
pgvector is an open-source Postgres extension for storing and querying vector embeddings directly in your database using standard SQL.
Key capabilities:
- Vector storage: Store embeddings alongside your existing data as a native Postgres column type.
- Similarity search: Query by cosine similarity, L2 distance, or inner product.
- HNSW and IVFFlat indexes: Scale approximate nearest-neighbor search to large datasets.
EDB Agent Governance
EDB Agent Governance is a standalone application that audits and governs how AI agents interact with your Postgres data. It connects to your HM-managed clusters or standalone Loki instances to reconstruct agent sessions from Postgres query logs.
Key capabilities:
- Reconstructing agent sessions: Inspect every SQL statement an AI agent executed, with full execution metadata.
- Auditing across multiple data sources: Connect to several HM or Loki sources at once and audit clusters across all of them from one place.
- Incremental sync: Session data is cached and synced incrementally, so audit views load quickly without re-querying the full log history on every visit.