Post

Similarity search is a leaky abstraction for structured knowledge. Instead of hoping top-k vector chunks have the right context, give your agent a purpose-steered outline via `initial_context` and read exact Markdown paths. Architecture in the reply 👇

21/10/26, 5:15 pm1 repost2 likes

Alex Patterson

@codercat.dev

The two-step retrieval pattern (`initial_context` outline in memory + explicit `knowledge_base_read` calls) is way more predictable than hoping top-k vector chunks line up.

CodingCat.dev

@codingcat.dev

We break down the build-time compilation pipeline, GROQ projections, and Context MCP tool mechanics in the full video and blog post: Video: https://youtu.be/jm5Vq60qYTk Article: https://codingcat.dev/post/sanity-knowledge-bases