Your Snowflake data is ready. Your AI agents aren’t — yet.

AutoGraph is an automatic knowledge-graph engine that builds AI-powered knowledge graphs from documents without manual setup. Users launch AutoGraph and create a project by naming it and uploading a document folder for automatic knowledge extraction. The system supports multiple chat and embedding models and is provider-agnostic, working with preferred models and API keys including OpenAI. A continuous slider adjusts complexity from 0% to 100%, with lower levels optimizing for speed and efficiency while higher complexity produces richer graphs with more entities, relationships, broader context, and multi-layer embeddings. Users set the number of parallel importers to control concurrent graph construction and accelerate data processing. AutoGraph embeds documents, identifies similarity relationships, and discovers natural domains within datasets. The strategizer generates custom ontologies for each domain and determines graph construction methods. In the example shown, 114 source documents were processed, resulting in 2 distinct domains with a FullGraphRAG strategy assigned to each. The system imports analyzed data into the knowledge graph, populating it with entities, relationships, and insights. The graph structure can be visualized in the Graphs section. Modules represent the top level of the graph, with documents grouped based on projects, departments, domains, or other organizational structures. AutoGraph automatically identifies domains within modules, each with its own RAG strategy generated by the Strategizer. Source documents are visible within each domain. The knowledge graph ontology shows how source documents are split into chunks, with entities extracted from each chunk. Users can trace from documents to chunks to individual entities. The retriever enables context-aware question answering. Three retrieval strategies are available: Local with Deep Search provides comprehensive answers with longer processing time, Unified delivers instant concise responses, and Global handles high-level questions about overarching themes and patterns. Users can specify response instructions to control answer style, include metadata, toggle caching, and choose whether to show citations. The system returns answers with detailed metadata for transparency and observability. The underlying data structure includes collections storing the Corpus Graph and Knowledge Graph. The RAGs collection stores retrieval strategies, with each domain having its own RAG profile including the automatically generated ontology and ingestion configurations. The Similarities edge collection shows how documents connect semantically, with similarity scores and rankings between related documents.