January 6, 2026
Fast, furious, and fact-checked
Show HN: Mantic.sh – Search 480k files in 0.46s without embeddings
Lightning-fast local code search sparks cost drama—devs demand receipts
TLDR: Mantic claims lightning-fast local code search without costly cloud databases, boasting 0.46s over 480k files. The community loves the speed but challenges the price math, the “cognitive” branding, and MCP claims, demanding clearer docs and features before crowning it the search king.
Mantic dropped into Show HN like a mic: “Search 480k files in 0.46s, no embeddings, no cloud,” promising sub‑half‑second results and local privacy. It even shipped a fresh fix for a scoring bug in v1.0.12 and claims it beats vector databases (the math-heavy tools that turn text into numbers) by 11–22x. Cue the crowd: instant split between speed hype and spreadsheet skepticism.
One camp cheered the “fast and local” angle for AI coding tools like Claude and Cursor, while others slammed Mantic’s cost chart as too good to be true. User oooyay called the price comparison “cynical” and asked for a breakdown, basically yelling show the receipts. esafak poked at the branding, questioning what “cognitive search without embeddings” even means, before dubbing it “structural search”—aka it guesses relevance from file names and paths, not content. Then came feature snags: raylad flagged missing ignore patterns, warthog wanted a deeper explainer, and vault wondered if Mantic was really an MCP tool or just wearing the merch. The vibe? Embeddings bros vs. heuristics hipsters, with jokes about “0.46s to drama.” Fans want speed and privacy; skeptics want docs, definitions, and fewer bold claims. The showdown isn’t over.
Key Points
- •Mantic provides sub-500ms structural code search across large codebases without embeddings, vector databases, or external dependencies.
- •v1.0.12 fixes a missing high-confidence scoring path, ensures proper BrainInspiredScorer scoring for all queries, and resolves undefined scoredFiles fallback issues from v1.0.10.
- •Benchmarks show 0.46s retrieval on Chromium (480k files) and 0.32s on Cal.com, outperforming vector search by 2.7x to 22x.
- •The system uses git ls-files for fast enumeration and a pipeline (Intent Analyzer → Brain Scorer → File Classifier → Impact Analyzer) with structural scoring and confidence metrics.
- •Mantic runs locally for privacy, can reduce token usage by up to 63%, integrates via MCP with tools like Claude Desktop and Cursor, and offers CLI options and editor installations.