Fetching from the wire…
OSS2026-07-30 · source-backed
HKUDS/LightRAG (EMNLP 2025), microsoft/graphrag, and milvus-io/milvus remain the three anchors of graph and vector retrieval, all clustered within 10K stars of each other. None has broken away, which is itself the signal: graph RAG hasn't picked a winner after eighteen months. If you're choosing a knowledge-graph retrieval backend today you're still picking among three viable, actively maintained options rather than accepting a default. That's more work for you and probably healthier for the ecosystem.
Each link below shares sources, entities, or timing with this story.
From HKUDS (the LightRAG lab), OpenHarness (Python, ~13,800 stars, v0.1.9) packages tool-use, skills, persistent memory, and multi-agent coordination, plus a personal assistant "ohmo" that talks to Slack and Feishu. It ships a 43-plus tool ecosystem with MCP support and unifie...
HKUDS/DeepTutor v1.5.12 (35,558 stars) rebuilt its web-search layer onto a single SEARCH_PROVIDERS spec table after the list had been copy-pasted into backend registry, runtime config, settings router, CLI wizard, frontend catalog, i18n and tests, then drifted. Consolidation e...
The lab behind LightRAG (37,823 stars) and nanobot (45,882) landed correlation-regime detection plus a backtest/data/live-safety correctness pass on July 17, then Binance crypto fallback and parallel-execution fixes on July 18. Be skeptical of any LLM trading agent's claims. B...
Built by Bingxi Zhao in Chao Huang's HKUDS lab, it runs Chat, Quiz, Research, Visualize, Solve and Mastery Path on one unified agent loop: "you switch the objective, not the engine, and context moves with the learner." The memory design is the transferable idea. L1 is a worksp...
Lifelong personalized tutoring built on the lab's own retrieval stack, targeting persistent learner models rather than per-session Q&A. One academic group with three separately successful agent projects is its own signal about where usable agent research is currently coming fr...
claude-context (11.8K stars) is an MCP server that indexes your whole codebase into Milvus and exposes hybrid BM25 + dense-vector search to Claude Code and any MCP agent. Ask "find functions that handle user authentication" instead of grepping. Zilliz reports ~40% token reduct...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.