Skip to main content
Context engineering is the discipline of designing and optimizing the information pipeline that feeds your AI agents. It encompasses everything from how data enters your system, how it gets enriched with structure and meaning, and how the right pieces are retrieved at query time to produce accurate, grounded responses. Airia provides a complete context engineering pipeline — from connecting your enterprise data to delivering precisely the right context to any LLM, through any interface.

​
The Airia Context Pipeline

Every piece of knowledge your agents use flows through four stages: 
Connect → Process → Enrich → Retrieve
Connect — Bring your data in from 20+ enterprise sources (SharePoint, Google Drive, Confluence, S3, and more) with real-time sync and permission enforcement. Process — Documents are parsed, chunked, and embedded into vector representations. Choose your PDF parser, enable image scanning, select your embedding model, and configure your vector database. Enrich — Optionally extract structured knowledge from your documents. Knowledge Graph Extraction identifies entities and relationships, creating a graph layer on top of your vector store that dramatically improves retrieval quality for complex queries. Retrieve — Search your knowledge base using semantic search, keyword search, hybrid combinations, or agentic multi-hop retrieval via MCP. Add reranking for precision. Let your agents decide dynamically what to search, when, and how many times.

​
Choosing a Retrieval Pattern

Airia supports three retrieval patterns. Choose based on your use case:
PatternHow it worksBest for
Data Search StepSingle-hop, embedding-based search. The full user query is used to find matching chunks in one pass.Simple Q&A, batch processing, predictable queries. Fast and low-cost.
MCP Multi-Hop RetrievalMulti-hop agentic retrieval via the Airia Datasource MCP Server. The LLM autonomously decides which sources to query, which tools to use, and how many searches to run.Complex questions, conversational agents, multi-source reasoning, accuracy-critical workflows.
Text-to-SQLTranslates natural language into SQL queries against structured data (CSV, XLSX).Numerical analysis, tabular data, structured reporting.
These patterns can be combined in a single agent. For example, an agent might use a Data Search Step for fast initial lookup and a MCP Multi-Hop Retrieval for deeper follow-up reasoning.

​
What’s in This Section

PageWhat you’ll learn
Connecting Data SourcesHow to connect enterprise sources, supported formats, sync scheduling, permissions
Ingestion SettingsPDF parsers, image scanning, embedding models, vector database configuration
Knowledge Graph ExtractionIndustry presets, custom entity types, how Graph RAG works
Custom Knowledge GraphsBuilding runtime graphs with Cypher queries
Retrieval MethodsData Search Step, MCP Multi-Hop Retrieval (MCP), Text-to-SQL, configuration
Hybrid Search and RerankingSemantic vs keyword search, fusion algorithms, reranker models
Graph-Enhanced RetrievalHow knowledge graphs boost retrieval quality

​
Guides