Extract normalized triplets from sources into graph-backed semantic memory.
npx skills add https://github.com/thistleknot/skills --skill agentic-kg-memory
Please help me install this Agent Skill. Skill: agentic_kg_memory Source: https://github.com/thistleknot/skills/tree/main/agentic_kg_memory Command: npx skills add https://github.com/thistleknot/skills --skill agentic-kg-memory
It helps you turn scattered source material into graph-backed semantic memory, retrieve the most relevant evidence for a question, and continuously update conclusions as new evidence arrives.
Run agentic_kg_memory to ingest your sources, build the triplet and page layers, and retrieve grounded throughlines for your next question.
High-intent search queries and answers about installing and using this skill.
To build a knowledge graph from sources, you extract normalized triplets with polarity and inference type, storing them as typed edges with searchable dense and sparse surfaces for retrieval.
Updating semantic memory involves maintaining updateable throughlines where conclusions are identified by premise sets, allowing you to revise, reinforce, weaken, or supersede them as evidence changes.
Evidence grounding uses BM25 and vector embedding retrieval to narrow candidate sources via sparse and dense similarity, optionally routing structural queries through the knowledge graph to synthesize answers.
Yes, graph-backed memory requires a SQLite or Chroma-backed storage setup to maintain the triplet and page layers necessary for query-time retrieval and durable throughline management.
Graph-backed memory applies to both semantic or fuzzy questions and graph-structured evidence lookup by combining BM25 vector similarity narrowing with knowledge graph routing.
Use throughlines for knowledge accumulation when you need durable abductive conclusions whose identity is the premise set, enabling you to revise or supersede them as new evidence arrives.
patricio0312revConfigure Pinecone, Chroma, pgvector, and Qdrant vector databases for semantic search.
JCETools-PetraDesign RAG pipelines with chunking, embedding, retrieval, and evaluation metrics.
qdrantConstruct Qdrant hybrid search prefetch queries combining BM25 with dense vectors.
ionmidoriBuild RAG systems with vector databases and semantic search.
