zvec-grep: Evidence Routing for Workspace Search

Overview

zvec-grep routes workspace questions to the evidence they require: semantic or hybrid indexed retrieval discovers unknown wording and locations, ranked lexical search supplies anchors, and managed ripgrep checks exact occurrences. The series follows that path from scanner scope and structural extraction through adaptive recall, rank fusion, freshness, execution modes, model compatibility, and remote-transfer authorization. Its measured benefit is conditional: a first-party BrowseComp-Plus study reported similar accuracy with lower average tokens, tool calls, and agent time under one fixed corpus and configuration. Adoption fits repeated cross-file discovery when supported formats, current index state, model, privacy, and setup costs are acceptable; exact search remains stronger for exhaustive claims.

Learning outcomes

Use indexed retrieval to discover unknown wording or locations; use managed ripgrep when exact coverage matters.
A zvec-grep result is a ranked, source-linked sample, not exhaustive proof.
Scope, extraction structure, freshness, model compatibility, and truncation shape what the index can support.
BrowseComp-Plus reported lower search resources with similar accuracy, but only for its fixed first-party setup.

Keywords

zvec-grepDiscoveryRetrievalripgrepIndexingEmbeddingsFreshnessMCPPrivacyEvidence

Chapters

zvec-grep: Discovery vs Exact Search

Why one interface keeps ranked and exhaustive routes distinct

8 min · Why one interface keeps ranked and exhaustive routes distinct

8 min

zvec-grep: Choosing a Search Route

Match anchors, completeness, and cross-file uncertainty

9 min · Match anchors, completeness, and cross-file uncertainty

9 min

zvec-grep: How a Workspace Becomes Searchable

Scope, structure, chunks, and model limits shape discovery

9 min · Scope, structure, chunks, and model limits shape discovery

9 min

zvec-grep: How Ranked Evidence Is Assembled

From lexical and vector candidates to source-linked context

8 min · From lexical and vector candidates to source-linked context

8 min

zvec-grep: Tracing a Cross-File Question

From behavioral clue to verified source relationship

8 min · From behavioral clue to verified source relationship

8 min

zvec-grep: Freshness When the Index Changes

States, refresh modes, and diagnostics for current evidence

10 min · States, refresh modes, and diagnostics for current evidence

10 min

zvec-grep: Choosing an Embedding Model

Balance workload fit, privacy, compatibility, and cost

10 min · Balance workload fit, privacy, compatibility, and cost

10 min

zvec-grep: What the Benchmark Actually Shows

Measured efficiency gains, setup costs, and evidence limits

11 min · Measured efficiency gains, setup costs, and evidence limits

11 min

zvec-grep: When to Adopt Indexed Discovery

A conditional rule for scope, trust, freshness, and cost

12 min · A conditional rule for scope, trust, freshness, and cost

12 min

Preview every chapter · Unlock full text, audio, and video from chapter 4

Plans

Ch.1-3 FreeCh.4-9 Premium

FAQ

When should I use zvec-grep instead of ripgrep?

Use zvec-grep for unknown wording, locations, or cross-file relationships, and use ripgrep for known anchors or exhaustive coverage.

Is a top zvec-grep result proof that no other match exists?

No. A top zvec-grep result is a ranked sample from indexed scope, so completeness claims still require current-source reading or managed ripgrep.

What determines whether a workspace is searchable with zvec-grep?

A workspace is searchable through zvec-grep only to the extent that scope rules admit its files, extraction preserves useful structure, and the model can embed the resulting fragments. Unsupported formats, fallback extraction, and truncation narrow the evidence.

When is zvec-grep worth adopting?

zvec-grep is worth adopting when repeated ambiguous or cross-file searches justify index setup and maintenance, supported formats and freshness are adequate, and model and privacy conditions are acceptable.