he build-index¶

Build or rebuild the vector search index for a knowledge abstract.

Synopsis¶

he build-index KA_PATH [OPTIONS]

Arguments¶

Argument	Description
`KA_PATH`	Path to knowledge abstract directory

Options¶

Option	Short	Description
`--force`	`-f`	Force rebuild even if index exists

Description¶

Builds a vector search index for semantic search and chat functionality:

Reads all entities/relations — From the knowledge abstract data
Generates embeddings — Using the configured embedding model
Builds FAISS index — For fast similarity search
Saves to disk — In the index/ subdirectory

Required for: he search and he talk commands.

Examples¶

Build Index¶

he build-index ./output/

Output:

Template: general/biography_graph
Language: en

Success! Index built for ./output/

Now you can:
  he search ./output/ "keyword"  # Semantic search
  he talk ./output/ -i           # Interactive chat

Force Rebuild¶

If the index is corrupted or you want to rebuild:

he build-index ./output/ -f

When to Build Index¶

After Parse (Default)¶

By default, he parse builds the index automatically:

he parse doc.md -t general/biography_graph -o ./ka/ -l en
# Index is built automatically

Skip with --no-index if you don't need search/chat:

he parse doc.md -t general/biography_graph -o ./ka/ -l en --no-index

After Feed¶

Always rebuild after feeding new documents:

he feed ./ka/ new_doc.md
he build-index ./ka/

After Manual Changes¶

If you modify data.json manually:

he build-index ./ka/ -f

Index Storage¶

The index is stored in the knowledge abstract directory:

./ka/
├── data.json
├── metadata.json
└── index/              # Index directory
    ├── index.faiss     # FAISS vector index
    └── docstore.json   # Document store mapping

Performance¶

Build Time¶

Knowledge Abstract Size	Approximate Build Time
Small (< 100 items)	< 5 seconds
Medium (100-1000)	5-30 seconds
Large (1000+)	30+ seconds

Search Speed¶

Once built, searches are fast:

he search ./ka/ "query"  # Typically < 1 second

Best Practices¶

Build after feeding — Index becomes stale after he feed
Use --no-index for batch — Build once after all parsing
Force rebuild if issues — Use -f if search returns unexpected results
Backup large indices — The index/ directory can be large

Batch Workflow¶

For processing multiple documents efficiently:

# Parse all without building index
he parse doc1.md -t general/biography_graph -o ./ka/ -l en --no-index
he feed ./ka/ doc2.md
he feed ./ka/ doc3.md

# Build index once at the end
he build-index ./ka/

# Now ready for search/chat
he search ./ka/ "query"
he talk ./ka/ -q "question"

Troubleshooting¶

"Index already exists"¶

Use -f to force rebuild:

he build-index ./ka/ -f

"Failed to build index"¶

Check: 1. Knowledge base has data: he info ./ka/ 2. API key is configured: he config show 3. Sufficient disk space for index

Search still doesn't work¶

Try force rebuild:

he build-index ./ka/ -f