Agent description
Agent for ingesting and querying structured and unstructured content. Accepts PDF, CSV, JSON, and TXT files, extracts text and metadata, normalizes the content, and imports it into a semantic graph to enable complex queries over informational links.
Key features
- Supported formats: PDF (with optional OCR), CSV, JSON, TXT.
- Extraction: parsing of tabular structures, extraction of text and metadata (author, timestamp, column names, etc.).
- Cleaning and normalization: noise removal, tokenization, segmentation into semantic chunks.
- Enrichment: extraction of entities, relationships, and semantic annotations to create nodes and edges.
- Graph ingestion: mapping entities to nodes, creating semantic edges, and storing them in the graph backend.
- Graph queries: searches by entity, paths between nodes, pattern matching and aggregations; support for textual and structured queries.
- Traceability: preserving references to original documents and text blocks to reconstruct context.
Benefits
- Transforms diverse document formats into a queryable graph representation.
- Enables semantic analysis and discovery of hidden relationships across heterogeneous data.
- Eases integration with existing pipelines (e.g., kgrag-store, MCP server) for advanced search and reasoning.
Example workflow
- File reception → 2. Extraction and normalization → 3. Entity/relationship extraction → 4. Graph ingestion → 5. Execute queries and retrieve results
Dependencies
kgrag-store: Core data storage and graph management library.memory-agent: A Python library for advanced memory management in AI agent applicationskgrag-mcp-server: server that implements the Model Context Protocol (MCP) for managing, ingesting, and querying structured and unstructured data