No More Data Silos: Semantic Layer Means Agents Can Query Databases Without ETL
AWS has partnered with Stardog to launch a new semantic layer architecture that lets agentic AI systems query Amazon Aurora and Amazon Redshift databases directly, bypassing the traditional extract, transform, and load (ETL) pipelines. According to an AWS Machine Learning blog post, the integration pairs Stardog’s Semantic AI Application with Amazon Bedrock AgentCore to create a unified knowledge graph that agentic workflows can query in natural language — effectively giving AI agents a single, canonical understanding of data spread across multiple cloud databases.
The announcement comes at a time when enterprise adoption of agentic AI is accelerating, but enterprises remain stuck with fragmented data architectures that often require weeks of ETL work before AI models can access relevant information. By eliminating ETL, the Stardog-on-AWS stack cuts down the time from data ingestion to actionable AI query from weeks to minutes.
What Was Announced
Stardog, which specializes in knowledge graph technology for enterprise data, has built its Semantic AI Application on top of Amazon Aurora (a relational database) and Amazon Redshift (a cloud data warehouse). The deployment runs on AWS compute services including Amazon Elastic Kubernetes Service (Amazon EKS), Amazon Elastic Container Service (Amazon ECS), and AWS Lambda. Specifically, developers can run a Strands Agents agent — Stardog’s agentic framework — on Amazon Bedrock AgentCore, which then queries the semantic layer to answer complex customer 360 questions across both sources without moving or transforming data.
Under the hood, Stardog maps the schema of each database into a unified ontology. When an agent asks a question like “Show me all customers who upgraded their plan in the last 30 days and their support ticket history,” the semantic layer translates that into the appropriate SQL or SPARQL queries against Aurora and Redshift simultaneously.
Why It Matters for AI Developers
For developers building agentic systems, this is a pragmatic shift away from the “one vector database to rule them all” mindset. The reality is that most enterprise data lives in transactional databases (Aurora) and analytical warehouses (Redshift). Trying to force that data into a single vector store often loses relational context. “Semantic layers preserve the relationships between entities — customer, order, product — which is exactly what agentic AI needs to reason about,” the blog post explains.
Stardog’s semantic layer adds a crucial missing piece: ontology-driven reasoning. While Large Language Models (LLMs) can generate SQL, they often produce hallucinated or incorrect queries against complex schemas. A semantic layer pre-validates the query logic against a formal ontology, reducing errors and improving trustworthiness.
- No ETL delays — Data remains in its original store; the semantic layer sits on top as a live view.
- Multi-source joins — Agents can query across Aurora and Redshift in a single request, enabling richer customer 360 analyses.
- Scalable compute — Runs on EKS, ECS, or Lambda, so it fits existing DevOps pipelines without provisioning new infrastructure.
Business Implications: Faster AI ROI, Lower Data Engineering Costs
From a business perspective, the biggest win is speed of AI implementation. McKinsey research suggests that data preparation accounts for up to 80% of the time spent on AI projects. By eliminating ETL for semantic queries, companies can reduce the time from idea to production agent by weeks. “An enterprise with a typical Aurora and Redshift setup can now have an agent answering customer support questions in hours, not months,” the announcement notes.
Additionally, the architecture reduces the need for specialized data engineering teams to build and maintain ETL pipelines for agentic workloads. Existing SQL-skilled developers can define the ontology once and let Stardog handle the query translation.
How It Works: A Developer’s Quick-Start View
Developers familiar with AWS can deploy the stack in three high-level steps:
- Deploy Stardog on AWS using EKS, ECS, or Lambda via AWS Marketplace or CloudFormation templates.
- Connect to Amazon Aurora and Amazon Redshift by pointing Stardog to the database endpoints.
- Create an ontology that maps your business entities (e.g., Customer, Order, Product) to the underlying table schemas.
- Deploy a Strands agent on Amazon Bedrock AgentCore that uses the Stardog semantic layer as its knowledge base for queries.
The agent then uses function calling to translate natural language requests into validated semantic queries. No custom middleware or data pipelines are required.
Competitive and Industry Context
The move puts AWS in direct competition with offerings like Google’s Cortex Agent Builder (which uses Vertex AI and BigQuery) and Microsoft’s Azure OpenAI + Semantic Kernel stack. However, AWS’s bet on Stardog’s knowledge graph approach — rather than pure vector search — signals a belief that relational reasoning is essential for enterprise-grade agentic AI. While vector databases like Pinecone or Weaviate excel at similarity search, they struggle with multi-hop reasoning (e.g., “Find customers from California who bought product X and later returned it”). Semantic layers handle this natively.
For startups and smaller teams, this is also interesting because the solution runs on Lambda — meaning a team can start with a serverless semantic layer and scale up without upfront infrastructure costs.
What Comes Next
The Stardog-and-AWS integration is available now through the AWS Marketplace. Developers interested in testing should look at the Stardog Strands Agents GitHub repository for sample ontologies and example agent definitions. AWS also plans to release a workshop in the coming weeks that walks through setting up a customer 360 agent from scratch.
If this approach gains traction, we may see more databases offering native semantic layer protocols, reducing the need for third-party tools. For now, however, the combination of Stardog’s knowledge graph and AWS’s compute and database services offers one of the most practical paths to agentic AI that actually works with real enterprise data.
Related: Context Graphs Turn Enterprise AI from Reactive to Proactive, New Research Shows
Related: OpenAI’s GPT-5.6 Now Powers Microsoft 365 Copilot: What Developers and Enterprises Need to Know
Source: AWS Machine Learning. This article was produced with AI assistance and reviewed for accuracy. Editorial standards.
