Feeding agents the right slices of large corpora: chunking libraries, parsing pipelines, search-engine libraries, full RAG engines, the two big frameworks, graph-RAG systems, a vectorless reasoning engine, code knowledge-graph retrieval, 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.
- Code-Graph-RAG - the MIT monorepo code-RAG system storing tree-sitter plus compiler facts in a Memgraph graph queried by an agent writing Cypher.
- Docling - IBM-origin document parser turning PDF, Office, audio, and video into structured DoclingDocuments for RAG and agent pipelines.
- GraphRAG - Microsoft Research’s entity-graph RAG pipeline with community summaries, now declared largely in maintenance mode by its own README.
- 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.
- LightRAG - the HKU lab’s MIT graph-RAG framework (dual-level retrieval, pluggable graph storage), the maintained carrier of the pattern Microsoft parked.
- LlamaIndex - the MIT data framework for retrieval pipelines, now the open arm of LlamaParse.
- open-codebase-index - the MIT self-hosted semantic code index (embeddings, BM25, call graph) served to OpenCode, Claude Code, Codex, Pi, and MCP clients.
- Orama - the Apache-2.0 embeddable search engine and RAG pipeline for JS/TS, 1.4M weekly installs on a stalled release train.
- PageIndex - Vectify AI’s vectorless reasoning-RAG engine (tree index per document, LLM tree search), the largest bet that the vector layer is skippable.
- R2R - SciPhi’s MIT agentic-RAG engine (multimodal ingestion, hybrid search, knowledge graphs, Deep Research API), two large Show HNs in 2024, dormant since late 2025 with its docs and company site gone.
- RAGFlow - InfiniFlow’s Apache-2.0 RAG engine (DeepDoc parsing, template chunking, citations, agentic retrieval), 92.1k stars, self-hosted or clouded, with a disruptive 1.0 Go rewrite.
- 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.
- Unstructured - the Apache-2.0 parsing incumbent Docling displaced, now funneling to a hosted Transform MCP server and quote-priced Pipelines.
Its members are compared on shared rows in the Retrieval Feature Matrix.
Changes #
- 2026-08-24 - Added LangChain.
- 2026-08-24 - Added LlamaIndex.
- 2026-08-24 - Added Semantic code search.
- 2026-08-24 - Added Tree-sitter chunking.
- 2026-09-16 - Added Chonkie.
- 2026-09-20 - Added Knowhere.
- 2026-09-27 - Added Docling.
- 2026-10-05 - Added open-codebase-index.
- 2026-10-06 - Added Orama.
- 2026-10-06 - Added RAGFlow.
- 2026-10-06 - Added Unstructured.
- 2026-10-07 - Added Code-Graph-RAG.
- 2026-10-07 - Added GraphRAG.
- 2026-10-07 - Added LightRAG.
- 2026-10-07 - Added PageIndex.
- 2026-10-11 - Added R2R.