Retrieval, context and memory

A useful answer depends on the evidence supplied to the model. Retrieval chooses material, context engineering assembles the current working set, and memory keeps selected information available for future tasks.

Begin with the retrieval pipeline and the distinction between embeddings, rerankers and classifiers. Then study context budgets and memory updates. QMD, zvec-grep and Basic Memory are implementations of different parts of this work, not mandatory stages in one stack.

Reading Order

Step	Article	What it explains
1	Retrieval Pipelines: Chunking, Recall, Reranking, and Citations	Turn documents into traceable evidence and identify what recall, ranking, and context assembly can lose.
2	Embeddings, Rerankers, and Classifiers: Similarity, Relevance, and Decisions	Distinguish three model roles by their inputs, outputs, and training objectives rather than backbone names.
3	Context Engineering for Agents	How to assemble a small, current working set while preserving provenance and durable state outside the model window.
4	Agent Memory: Retention, Revision and Expiry	Choose what an AI assistant should remember and retrieve, without confusing stored notes, search results, and authority.
5	QMD: Search Markdown Notes Locally	Find Markdown notes with keyword, semantic, or hybrid search, then read the source behind each result.
6	zvec-grep: Search Code and Documents by Meaning	Combine semantic discovery and keyword search to locate relevant workspace files, then verify the current source.
7	Basic Memory: Linked Markdown Notes for AI Assistants	Store, connect, and retrieve Markdown notes across conversations while keeping recorded decisions separate from permissions.

Use What You Read

For a missed answer, locate the failure: absent source, missed candidate, poor ranking, lost context or unsupported generation. Preserve source references and use retrieval/generation evaluation to test each boundary.

Return to the AI reading paths to choose a neighboring topic.

My reading · Opened notes are remembered in this browser.