Doctree FAQs

Question 1

What are the performance benefits of using Doctree?

Accepted Answer

Doctree offers fast indexing (e.g., 2-5s for 900 docs), low-latency search (5-30ms), and a minimal memory footprint (~25-50MB for 900 docs). Crucially, it significantly reduces LLM token costs by providing highly precise content (2K-8K tokens) compared to noisy vector RAG chunks.

Question 2

What are the core features of Doctree for document access?

Accepted Answer

Doctree offers BM25 keyword search with facet filtering, hierarchical document outline navigation, precise section content retrieval by node ID, and efficient full and incremental indexing. It also supports configurable ranking, frontmatter, and glossary expansion.

Question 3

What is Doctree and how does it help AI agents?

Accepted Answer

Doctree provides AI agents with structured, outline-guided access to markdown documents. It enables precise information retrieval by allowing agents to search with BM25 and navigate a document's hierarchical outline, reasoning about which sections are most relevant.

Question 4

How does Doctree differ from traditional RAG (Retrieval-Augmented Generation)?

Accepted Answer

Unlike traditional RAG that often provides a 'bag of paragraphs,' Doctree gives agents a navigable table of contents and a precise search engine. This allows agents to reason over the document structure and retrieve exact sections, leading to more accurate responses and reduced token costs, without needing vector databases or embeddings.

Question 5

Does Doctree use vector databases or embeddings?

Accepted Answer

No, Doctree operates without vector databases, embeddings, or LLM calls at index time. It leverages BM25 for keyword search and a native markdown parser to construct a hierarchical outline, focusing on structured and precise access.

Doctree

Doctree

主な機能

ユースケース

主な機能

ユースケース