Knowledge Share
Technical articles, tutorials, and insights
Codebase Knowledge Base Series (05): Vector Retrieval vs Knowledge Graph — Adding a Call Graph Didn't Help?
Vector retrieval (Recall@5=0.958) vs graph-augmented retrieval (seed_k=3, BFS 2 hops): identical overall scores, but different failure modes. Graph expansion fixes Q8 (calculate_order_total reached via process_checkout in 2 hops) but breaks Q1 (expanded candidate pool crowds out verify_password). Conclusion: naive graph traversal is double-edged; the real engineering value is encoding structural information into embedding content, not bolting on graph traversal at retrieval time.
Codebase Knowledge Base Series (01): Technical Landscape — Why Code Understanding Is 10x Harder Than Document Retrieval
A codebase isn't an upgraded document store — it's a fundamentally different knowledge type. Four understanding layers (syntactic/semantic/architectural/intent) map to four retrieval requirements. Traditional grep, vector search, AST symbol indexing, and call graph retrieval each have distinct capability limits. This article builds the complete framework for codebase knowledge systems.
Open Source Project of the Day (#155): Cognee — Persistent Cross-Session Memory for AI Agents
Cognee is an open-source AI agent memory platform that combines vector retrieval, knowledge graphs, and cognitive science methods into a unified persistent memory layer. Four core operations: remember/recall/forget/improve. BEAM benchmark: 0.79 at 100K tokens, beating prior SOTA. 27.4k Stars, Apache-2.0.
Open Source Project of the Day (#152): SAG — Multi-Hop RAG Retrieval via SQL JOINs Instead of PageRank
Zleap-AI's Structured Agentic Graph RAG framework extracts each document chunk into one semantic event and multiple entities, then uses SQL JOINs at query time for multi-hop expansion — no global graph pre-built, no PageRank decay. MuSiQue Recall@5: HippoRAG 2's 65.1% → SAG's 80.0%. MIT license with MCP integration.
Open Source Project of the Day (#149): RAG-Anything — Making Images, Tables, and Equations First-Class Citizens in RAG
HKUDS's all-in-one multimodal RAG framework built on LightRAG, treating images, tables, and LaTeX equations as first-class knowledge graph entities rather than degrading them to plain text. DocBench: 63.4% vs MMGraphRAG 61.0% and LightRAG 58.4%. Supports PDF/DOCX/PPTX and more. 21.9k Stars, MIT license.
Open Source Project of the Day (#147): HyperGraphRAG — N-ary Relations via Hyperedges, the Third-Generation RAG Paradigm
Official NeurIPS 2025 implementation of the first hypergraph-based RAG system. Replaces knowledge graph binary edges with hyperedges that natively represent multi-entity (N-ary) relationships. Outperforms GraphRAG and LightRAG across medicine, agriculture, computer science, and law. MIT license.
Open Source Project of the Day (#135): codebase-memory-mcp — A Knowledge Graph That Gives AI Agents Structural Memory of Your Codebase
A pure-C, zero-dependency MCP server that indexes codebases into a persistent knowledge graph. Reduces code exploration from 412,000 tokens to 3,400 tokens (120x). Supports 158 languages, sub-millisecond Cypher queries, and 14 MCP tools. 5.4k Stars, MIT license.
One Open Source Project a Day (No. 111): Understand Anything - The AI Engine That Turns Any Codebase Into an Explorable Knowledge Graph
Understand Anything is a Claude Code plugin that converts codebases, knowledge bases, and documentation into interactive knowledge graphs. Its hybrid Tree-sitter + LLM architecture separates deterministic structural extraction from semantic understanding, using a 5-agent pipeline to make any codebase navigable by natural language. 26.5k Stars — one of the most-watched code comprehension tools of 2026.
One Open Source Project a Day (No. 107): CodeGraph — Pre-Index Your Codebase for AI Agents, Save 35% Cost and 70% Tool Calls
CodeGraph is a local semantic code knowledge graph tool that pre-parses codebases with tree-sitter into a SQLite index, then exposes it to AI coding agents like Claude Code and Cursor via MCP. Benchmarks across 7 real-world projects show an average 35% cost reduction, 70% fewer tool calls, and 49% speed improvement — on VS Code's large TypeScript repo, one task drops from 1.4M tokens to 393k. Completely local, no API keys required.
RAG Series (16): Graph RAG — Using Knowledge Graphs to Solve Multi-Hop Reasoning
Part 16 of the RAG series. Vector retrieval finds "similar documents" but is blind to entity relationships. Graph RAG builds a knowledge graph from documents and uses graph traversal instead of similarity search. 12 documents → 176 nodes / 139 edges. Results: context_precision jumps from 0.729 to 0.948 (+0.219), with an honest tradeoff: context_recall drops 0.062.
Open Source Project of the Day (Part 28): Graphiti - Building Real-Time Knowledge Graphs for AI Agents
A deep dive into Graphiti, getzep's open-source "temporally-aware" knowledge graph framework — supporting incremental updates, dual temporal models, and hybrid retrieval for AI Agent memory and context engineering in dynamic environments, compatible with Neo4j/FalkorDB/Kuzu/Neptune and MCP
Open Source Project of the Day (Part 44): GitNexus - Zero-Server Code Intelligence Engine for Building Codebase Knowledge Graphs for AI Agents
A deep dive into GitNexus, a zero-server code intelligence engine that runs entirely in your browser, creating knowledge graphs and providing deep codebase awareness to AI agents like Cursor, Claude Code, and Windsurf through the MCP protocol