knowledge

Knowledge Examples

Knowledge-enhanced AI agents examples.

Environment

export OPENAI_BASE_URL=xxx
export OPENAI_API_KEY=xxx
export MODEL_NAME=xxx

Common flags

Most examples accept:

-vectorstore inmemory|pgvector|tcvector|elasticsearch  # default: inmemory

Examples

chunking/

Offline text and interactive web viewers that show each Reader's default strategy, with advanced comparisons for FixedSizeChunking, RecursiveChunking, MarkdownChunking, and JSONChunking.

basic/

Basic example with file source and in-memory vector store.

contextual-retrieval/

Opt-in index-time contextual retrieval using a custom source wrapper. It keeps the original chunk content unchanged, caches generated context locally, and supports paired baseline / contextual trials. This is an evaluation example, not a default recommendation; see its README for evidence, security, and cost boundaries.

sources/

Different data source types:

Case	Description
file-source/	Load individual files (Markdown, PDF, DOCX, CSV, JSON)
directory-source/	Recursively load entire directory
url-source/	Fetch and parse web pages
auto-source/	Auto-detect source type from path

vectorstores/

Persistent vector storage options:

Case	Description	Extra Environment
chroma/	ChromaDB over the Chroma v2 REST API	CHROMA_URL, CHROMA_COLLECTION, CHROMA_API_KEY, CHROMA_TENANT, CHROMA_DATABASE
postgres/	PostgreSQL with pgvector extension	PGVECTOR_HOST, PGVECTOR_PORT, PGVECTOR_USER, PGVECTOR_PASSWORD, PGVECTOR_DATABASE, PGVECTOR_TABLE
elasticsearch/	Elasticsearch (v7/v8/v9)	ELASTICSEARCH_HOSTS, ELASTICSEARCH_USERNAME, ELASTICSEARCH_PASSWORD, ELASTICSEARCH_INDEX_NAME
tcvector/	Tencent VectorDB	TCVECTOR_URL, TCVECTOR_USERNAME, TCVECTOR_PASSWORD, TCVECTOR_COLLECTION
milvus/	Milvus vector database	MILVUS_ADDRESS, MILVUS_USERNAME, MILVUS_PASSWORD, MILVUS_DB_NAME, MILVUS_COLLECTION

features/

Advanced features:

Case	Description
agentic-filter/	LLM automatically generates metadata filter based on user query
metadata-filter/	Programmatic metadata filtering with AND/OR/NOT operations
management/	Dynamic source management: AddSource, RemoveSource, ReloadSource
batch-embedding/	Group documents into multi-input embedding requests with WithEmbeddingBatchSize