Retrieval

Feeding agents the right slices of large corpora: chunking libraries, parsing pipelines, the two big frameworks, and the patterns built on them.

Chonkie - the MIT chunking library (token, semantic, and neural chunkers) for RAG pipelines, its commercial arm dead and its maker pivoted to Feyn Labs.
Knowhere - Ontos AI’s structure-preserving document parsing and retrieval pipeline, hosted per page or self-hosted, its self-reported benchmark and near-empty HN footprint attached.
LangChain - the largest LLM framework, repositioned in 2026 as an agent engineering platform.
LlamaIndex - the MIT data framework for retrieval pipelines, now the open arm of LlamaParse.
Semantic code search - retrieval by meaning over embedded chunks, shipped as a workspace index.
Tree-sitter chunking - cutting files along syntax boundaries instead of fixed line counts.

Its members are compared on shared rows in the Retrieval Feature Matrix.

Changes
#