@layrx/search-indexer

Search Indexer for LayrX. Performs semantic and metadata-based retrieval over repository entities using embeddings from the Embedding Worker and metadata from the Knowledge Graph.

Architecture

User Query
     ↓
SearchIndexer (public API)
     ↓
SearchEngine (candidate loading + filtering)
     ↓
Embeddings (SQLite) + Memory (Knowledge Graph) + Files (recency)
     ↓
RankingEngine (multi-signal scoring)
     ↓
Search Results

| File | Responsibility | |------|----------------| | SearchIndexer.ts | Orchestrator and public API | | SearchEngine.ts | Candidate loading, filtering, search execution | | RankingEngine.ts | Extensible multi-factor ranking | | SimilarityCalculator.ts | Cosine similarity and text relevance helpers | | SearchFilters.ts | Zod validation and filter utilities | | SearchStatistics.ts | Query logs and aggregate metrics | | SearchTypes.ts | Shared types |

Prerequisites: Repository indexed → dependency graph → memory built → embeddings generated.

Search types

| Type | Description | |------|-------------| | semantic | Cosine similarity between query and stored embedding vectors | | exact | Exact name or path match | | prefix | Prefix/substring match on name or path | | hybrid | Weighted combination of all ranking factors (default) |

Ranking factors

Hybrid mode combines:

• Embedding similarity (45%) — vector cosine similarity
• Exact name match (25%) — case-insensitive name equality
• Path relevance (15%) — path segment overlap
• Graph relationships (10%) — relation count from memory_relations
• Recently modified (5%) — recency from files.modified_at

Weights are configurable via SearchIndexerOptions.rankingWeights.

Filters

• entityType — File, Function, Class, etc.
• directory — path prefix (e.g. src/components)
• fileExtension — e.g. .ts or ts
• repositoryId — scope to a repository

Public API

import { indexRepository } from '@layrx/repository';
import { buildDependencyGraph } from '@layrx/dependency-graph';
import { buildMemory } from '@layrx/memory';
import { generateAll } from '@layrx/embedding-worker';
import {
  search,
  searchByType,
  searchByPath,
  getSuggestions,
  getStatistics,
} from '@layrx/search-indexer';

await indexRepository('./my-repo');
await buildDependencyGraph('./my-repo');
buildMemory('./my-repo');
await generateAll('./my-repo');

const results = await search('Button', './my-repo', 'hybrid');
console.log(results.results);

await searchByType('Button', 'Function', './my-repo');
await searchByPath('Button', 'src', './my-repo');
getSuggestions('But', './my-repo');
getStatistics('./my-repo');

Result shape

Each result includes:

• entityId, entityType, name, path
• score (0–1)
• matchedReason — why the entity matched

Statistics

• totalIndexedEntities
• averageSearchTime
• cacheHitRate — placeholder (0 until AI Cache Manager)
• topSearchTypes
• lastIndexedAt

Logging

Search started → Candidates loaded → Ranking completed → Results returned → Errors

Schema

search_index_metadata

Per-repository indexed entity counts and timestamps.

search_query_log

Query history for statistics (query, search type, duration, result count).

Search flow

User Query → Load Memory + Embeddings → Apply Filters → Embed Query (semantic/hybrid)
→ Rank Candidates → Return Top Results → Log Statistics

Future improvements

• Approximate nearest neighbor (ANN) index for large vector sets
• External vector database sync (pgvector, Qdrant, Pinecone)
• Query result caching via AI Cache Manager
• Faceted search and result highlighting
• Cross-repository search

Tests

npm run test --workspace=@layrx/search-indexer

Do not use this package for caching, security scanning, or HTTP APIs — those belong to later features.