Knowledge graph memory
Skill Amey-Thakur/AI-SKILLS/skills/ai-memory-rag/knowledge-graph-memory
Store facts as entities and relationships when questions depend on connections rather than similarity. Use when retrieval must answer who relates to what rather than find similar text.From its SKILL.md
npx -y skills add Amey-Thakur/AI-SKILLS --skill knowledge-graph-memoryAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 5 stars5 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
1.9 KB, as published. Nobody here has run it
Knowledge graph memory
Vector retrieval finds text that resembles the query. Some questions are not about resemblance at all: who reports to whom, which service depends on which, what changed between these two things. Those are graph questions, and similarity search answers them badly.
Method
- Use a graph when relationships are the query. Traversal, dependency, and connection questions belong here; similarity and passage retrieval do not.
- Define the entity and relationship types deliberately. A schema with a handful of clear types is queryable; an open extraction producing hundreds of relation names is not.
- Resolve entities carefully. The same person or service named three ways becomes three nodes, and entity resolution is where graph quality is won or lost (see deduplication-queries).
- Keep provenance on every edge. Which document asserted this relationship and when, since graphs built by extraction contain errors that must be traceable.
- Combine graph with text retrieval. Traverse to find the relevant entities, then retrieve their supporting passages for the model to read (see hybrid-search).
- Update relationships as sources change. A stale edge is more misleading than stale prose because it reads as structured fact (see rag-freshness).
- Bound traversal depth. Deep queries return large subgraphs that exceed context and rarely improve the answer.
Boundaries
Graphs cost extraction and maintenance that a text index does not, and automated extraction is error-prone. They answer structural questions and are poor at nuance that lives in prose. A small corpus rarely justifies the construction effort.