Retrieval, context and memory

This group is about putting the right material in front of the model: finding the relevant passages in your documents, assembling the context for the current task, and deciding what persists across sessions. Each of the three local tools serves one need: QMD searches your own notes, zvec-grep finds files in code and documents by meaning, and Basic Memory lets an AI assistant read and write notes itself.

Reading Order

Step	Article	Your job
1	Retrieval Pipelines: Chunking, Recall, Reranking, and Citations	Turn documents into traceable evidence, and find what recall, ranking and context assembly each lose.
2	Embeddings, Rerankers, and Classifiers: Similarity, Relevance, and Decisions	Tell the three model roles apart by input, output and training objective, and pick the right one for each step.
3	Context Engineering for Agents	Assemble a small, current working set for the task, with provenance and durable state kept outside the window.
4	Agent Memory: Retention, Revision and Expiry	Decide what an AI assistant should remember, when to revise it and when it expires.
5	Vector Indexes: Exact Search, HNSW, and IVF	Compare exact search, HNSW and IVF, and measure recall against exact neighbours.
6	QMD: Search Markdown Notes Locally	Search your own Markdown notes by keyword, meaning or both, then read the source behind each hit.
7	zvec-grep: Search Code and Documents by Meaning	Find files in code and documents by meaning, then check the current source.
8	Basic Memory: Linked Markdown Notes for AI Assistants	Let an AI assistant read and write linked Markdown notes across conversations.
See also	Evaluating Retrieval and Generation: From Evidence to Correct Answers	Measure evidence hits, ranking, faithfulness and answer quality separately to find a RAG system’s bottleneck; this note belongs to the Evaluation and evidence group.

My reading