Crate reflex 

Reflex: Local-first, structure-aware code search engine

Reflex is a fast, deterministic code search tool designed specifically for AI coding agents. It provides structured results (symbols, spans, scopes) with sub-100ms latency by maintaining a lightweight, incremental cache in .reflex/.

Architecture

• Indexer: Scans code and builds trigram index; writes to cache
• Query Engine: Loads cache on demand; executes deterministic searches; parses symbols at runtime
• Cache: Memory-mapped storage for trigrams, content, and metadata

Example Usage

use reflex::{cache::CacheManager, indexer::Indexer, models::IndexConfig};

// Create and initialize index
let cache = CacheManager::new(".");
let config = IndexConfig::default();
let indexer = Indexer::new(cache, config);
let stats = indexer.index(".", false).unwrap();

println!("Indexed {} files", stats.total_files);

Re-exports

pub use cache::CacheManager;

pub use indexer::Indexer;

pub use models::Dependency;

pub use models::DependencyInfo;

pub use models::FileGroupedResult;

pub use models::ImportType;

pub use models::IndexConfig;

pub use models::IndexStats;

pub use models::IndexStatus;

pub use models::IndexWarning;

pub use models::IndexWarningDetails;

pub use models::IndexedFile;

pub use models::Language;

pub use models::MatchResult;

pub use models::QueryResponse;

pub use models::SearchResult;

pub use models::Span;

pub use models::SymbolKind;

pub use query::QueryEngine;

pub use query::QueryFilter;

pub use watcher::watch;

pub use watcher::WatchConfig;

Modules

ast_query

AST pattern matching using Tree-sitter queries

background_indexer

Background symbol indexer for transparent caching

cache

Cache management and memory-mapped I/O

cli

CLI argument parsing and command handlers

content_store

Content store for memory-mapped file access

context

Codebase context generation for AI prompts

dependency

Dependency tracking and graph analysis

formatter

Terminal output formatting for query results

git

Git repository utilities for branch tracking

indexer

Indexing engine for parsing source code

interactive

line_filter

Line-based pre-filtering to detect comments and string literals

mcp

MCP (Model Context Protocol) server implementation

models

Core data models for Reflex

output

User-facing output utilities for clean, colored terminal messages

parsers

Tree-sitter parsers for extracting symbols from source code

query

Query engine for searching indexed code

regex_trigrams

Extract literal sequences from regex patterns for trigram optimization

semantic

Semantic query generation using LLMs

symbol_cache

Symbol cache for storing parsed symbols

trigram

Trigram-based inverted index for fast full-text code search

watcher

File system watcher for automatic reindexing