wiki-retrieve

What problem does it solve?

It improves answer quality by finding the most relevant passages inside a large Obsidian “Compound Vault” at chunk-level granularity instead of relying on slower, page-level reads.

Core Features & Use Cases

• Contextual-prefix chunking: Builds paragraph-boundary chunks and contextual prefixes for each chunk to better match real questions to the right text.
• Hybrid retrieval pipeline: Uses BM25 to shortlist candidates and then applies dense cosine reranking to reorder results by semantic relevance.
• Feature-gated, safe integration: Detects whether the retrieval index is provisioned and otherwise cleanly falls back to the legacy hot→index→drill order.
• Privacy-aware egress control: Supports explicit consent gating for off-machine contextual-prefix generation, with an on-machine synthetic mode option.

Quick Start

Install retrieval indexing by running bash bin/setup-retrieve.sh.