Hypergraph is more powerful. Transform unstructured text into structured knowledge with LLMs. Graphs, hypergraphs, and spatio-temporal extractions — with one command.
CLI Workflow Guide
This guide walks you through a complete knowledge extraction workflow using real-world examples.
Scenario: Research Paper Analysis
Imagine you're a researcher who wants to extract and interact with knowledge from academic papers. Here's how to do it with Hyper-Extract.
Step 1: Prepare Your Document
Hyper-Extract supports .md and .txt files. Download the sample research paper below to follow this guide:
Save the file as transformer_paper.md in your working directory.
Step 2: Extract Knowledge
Use the parse command to extract knowledge:
he parse transformer_paper.md -t general/concept_graph -o ./transformer_kb/ -l en
What happens:
- The document is read and chunked (if long)
- LLM extracts entities and relationships
- Results are saved to
./transformer_kb/
Output structure:
./transformer_kb/
├── data.json # Extracted knowledge
├── metadata.json # Extraction metadata
└── index/ # Vector search index (if built)
Step 3: Explore the Knowledge
View Statistics
he info ./transformer_kb/
Output:
Knowledge Abstract Info
Path ./transformer_kb/
Template general/concept_graph
Language en
Created 2024-01-15 10:30:00
Updated 2024-01-15 10:30:00
Nodes 12
Edges 15
Index Built
Visualize
he show ./transformer_kb/

This opens an interactive graph in your browser showing:
- Nodes: Authors, concepts, models, metrics
- Edges: Relationships between them
Step 4: Search for Information
Semantic Search
Find information even if keywords don't match exactly:
he search ./transformer_kb/ "neural network architecture"
Results:
Found 3 result(s):
Result 1:
{
"name": "Transformer",
"type": "model",
"description": "A neural network architecture based solely on attention mechanisms"
}
Result 2:
{
"name": "Attention Mechanism",
"type": "concept",
"description": "Mechanism to draw global dependencies between sequences"
}
...
Specific Queries
he search ./transformer_kb/ "performance metrics" -n 5
Step 5: Chat with Your Knowledge
Single Question
he talk ./transformer_kb/ -q "What is the main contribution of this paper?"
Response:
The main contribution is the introduction of the Transformer architecture,
which achieves state-of-the-art results on machine translation tasks while
being more parallelizable and requiring significantly less training time than
recurrent or convolutional architectures.
Interactive Mode
he talk ./transformer_kb/ -i
Session:
Entering interactive mode. Type 'exit' or 'quit' to stop.
> Who are the authors?
The paper was authored by Ashish Vaswani, Noam Shazeer, Niki Parmar,
Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and
Illia Polosukhin.
> What BLEU score did they achieve?
The model achieved a BLEU score of 28.4 on the WMT 2014 English-to-German
translation task.
> exit
Goodbye!
Step 6: Expand Your Knowledge Abstract
Add Another Document
# Download another paper
curl -o bert_paper.md https://example.com/bert.md
# Add to existing knowledge abstract
he feed ./transformer_kb/ bert_paper.md
Verify the Update
he info ./transformer_kb/
Notice the increased node/edge count and updated timestamp.
Step 7: Rebuild Index (If Needed)
After adding documents, rebuild the search index:
he build-index ./transformer_kb/
Or force a complete rebuild:
he build-index ./transformer_kb/ -f
Complete Script Example
Here's a bash script that automates the entire workflow:
#!/bin/bash
# Configuration
INPUT_FILE="paper.md"
OUTPUT_DIR="./paper_kb/"
TEMPLATE="general/concept_graph"
LANGUAGE="en"
echo "=== Hyper-Extract Workflow ==="
echo
# Step 1: Parse
echo "Step 1: Extracting knowledge..."
he parse "$INPUT_FILE" -t "$TEMPLATE" -o "$OUTPUT_DIR" -l "$LANGUAGE"
echo
# Step 2: Info
echo "Step 2: Knowledge base info:"
he info "$OUTPUT_DIR"
echo
# Step 3: Search example
echo "Step 3: Sample search:"
he search "$OUTPUT_DIR" "main contributions" -n 2
echo
# Step 4: Open visualization
echo "Step 4: Opening visualization..."
he show "$OUTPUT_DIR"
echo "=== Workflow Complete ==="
Advanced: Batch Processing
Process multiple documents at once:
# Create output directories
mkdir -p ./ka/paper1 ./ka/paper2 ./ka/paper3
# Process each
he parse papers/paper1.md -t general/concept_graph -o ./ka/paper1/ -l en
he parse papers/paper2.md -t general/concept_graph -o ./ka/paper2/ -l en
he parse papers/paper3.md -t general/concept_graph -o ./ka/paper3/ -l en
# Or use a loop
for file in papers/*.md; do
name=$(basename "$file" .md)
he parse "$file" -t general/concept_graph -o "./ka/$name/" -l en
done
Common Patterns
Pattern 1: Continuous Knowledge Building
# Initial extraction
he parse initial_doc.md -t general/biography_graph -o ./ka/ -l en
# Weekly updates
he feed ./ka/ week1_update.md
he feed ./ka/ week2_update.md
he feed ./ka/ week3_update.md
# Monthly rebuild
he build-index ./ka/ -f
Pattern 2: Multi-Domain Project
# Technical documentation
he parse api_docs.md -t general/concept_graph -o ./project_kb/tech/ -l en
# Legal contracts
he parse contract.md -t legal/contract_obligation -o ./project_kb/legal/ -l en
# Financial reports
he parse q4_report.md -t finance/earnings_summary -o ./project_kb/finance/ -l en
Pattern 3: Compare Documents
# Extract two versions
he parse draft_v1.md -t general/concept_graph -o ./ka/v1/ -l en
he parse draft_v2.md -t general/concept_graph -o ./ka/v2/ -l en
# Compare via chat
he talk ./ka/v1/ -q "What are the main topics?"
he talk ./ka/v2/ -q "What are the main topics?"
Troubleshooting Tips
| Issue | Solution |
|---|---|
| Extraction is slow | Long documents are chunked; use --no-index to skip indexing during parse |
| Search returns nothing | Ensure index is built: he build-index ./ka/ |
| Template not found | List available: he list template |
| Out of memory | Reduce chunk size in config or process smaller documents |
Next Steps
- Learn about all CLI commands
- Explore the Template Library
- Read about choosing the right Auto-Type