Feed the graph with new knowledge
Drop documents or paste snippets to trigger the Graph-RAG ingestion flow. Monitor progress, inspect extracted triples, and capture logs without leaving the console.
Upload centre
Combine drag-and-drop uploads with raw text to seed the pipeline. Files stream as multipart payloads while pasted text is sent as JSON.
Drop files or click to browse
Supports .txt, .md, .pdf
Metadata travels with uploaded files and improves source scoring downstream.
Domain pack & controls
Select the active domain pack for tailored extraction rules. Switching is optimistic and synced with the API.
Need a new taxonomy? Extend graph-rag-infra packs and the console will pick it up automatically.
Disable this when you only want document and chunk ingestion without extracting or persisting graph triples.
When disabled, triples from a single context are retained for reasoning. Enable to require evidence from multiple contexts.
Advanced mode uses GPT-4o with 3-stage pipeline: entity extraction → triple extraction → ensemble refinement. Higher accuracy but slower and more expensive.
When enabled, chunks are processed concurrently. Speeds up long documents. Use 1–8 workers. Entity linking and graph boosting run in parallel.
Choose which document parsing service to use for uploaded files. Only applies when files are attached.
Enable to fail ingestion if protocol/legacy doc-tree building fails. Use this when tree-guided retrieval must always be available.
Pipeline telemetry
Rolling window metrics collected automatically from recent ingestion runs.
Run an ingest to populate telemetry. Metrics refresh automatically every few seconds.