Skip to content

README.md

Latest commit

 

History

History
131 lines (89 loc) · 4 KB

README.md

File metadata and controls

131 lines (89 loc) · 4 KB

@renseiai/agentfactory-code-intelligence

Code intelligence for AgentFactory agents — regex-based symbol extraction, BM25 search, incremental Merkle-tree indexing, PageRank repo maps, and memory deduplication.

Part of the AgentFactory monorepo.

Installation

npm install @renseiai/agentfactory-code-intelligence

Features

Module What it does
Parser Extracts symbols (functions, classes, types, interfaces) from TypeScript, Python, Go, and Rust using regex-based extractors
Search BM25-ranked code search with code-aware tokenization (camelCase splitting, stop-word removal) and exact/fuzzy match boosting
Indexing Merkle-tree change detection with git-compatible hashing — only re-indexes files that changed
Repo Map Builds a dependency graph from imports, ranks files by PageRank, and outputs an LLM-friendly repository map
Memory Two-tier deduplication: xxHash64 for exact matches, SimHash for near-duplicates with configurable Hamming distance

Usage

Symbol extraction

import { SymbolExtractor } from '@renseiai/agentfactory-code-intelligence'

const extractor = new SymbolExtractor()
const ast = extractor.extractFromSource(sourceCode, 'src/index.ts')

for (const symbol of ast.symbols) {
  console.log(`${symbol.kind}: ${symbol.name} (line ${symbol.line})`)
}

Code search

import { SearchEngine, SymbolExtractor } from '@renseiai/agentfactory-code-intelligence'

const extractor = new SymbolExtractor()
const engine = new SearchEngine()

// Extract symbols from your files, then build the index
const symbols = files.flatMap(f => extractor.extractFromSource(f.content, f.path).symbols)
engine.buildIndex(symbols)

const results = engine.search({ query: 'handleRequest', maxResults: 10 })

Incremental indexing

import { IncrementalIndexer, SymbolExtractor } from '@renseiai/agentfactory-code-intelligence'

const indexer = new IncrementalIndexer(new SymbolExtractor())

// Load previous index (if any)
await indexer.load(process.cwd())

// Index files — only changed files are re-processed
const files = new Map([['src/app.ts', sourceCode]])
const { changes, indexed, metadata } = await indexer.index(files)

console.log(`${changes.added.length} added, ${changes.modified.length} modified`)

// Persist for next run
await indexer.save(process.cwd())

Repository map

import { SymbolExtractor, RepoMapGenerator } from '@renseiai/agentfactory-code-intelligence'

const extractor = new SymbolExtractor()
const generator = new RepoMapGenerator()

const asts = files.map(f => extractor.extractFromSource(f.content, f.path))
const entries = generator.generate(asts, { maxFiles: 30 })

console.log(generator.format(entries))

Memory deduplication

import { DedupPipeline, InMemoryStore } from '@renseiai/agentfactory-code-intelligence'

const pipeline = new DedupPipeline(new InMemoryStore())

// Store content
await pipeline.storeContent('entry-1', 'some code block...')

// Check for duplicates
const result = await pipeline.check('some code block...')
// → { isDuplicate: true, matchType: 'exact', existingId: 'entry-1' }

Agent tool plugin

When used with AgentFactory's Claude provider, the package registers four in-process MCP tools:

Tool Description
af_code_search_symbols Search symbols by name with kind/language/file filtering
af_code_get_repo_map PageRank-ranked repo map of the most important files
af_code_search_code BM25 code search with code-aware tokenization
af_code_check_duplicate xxHash64 exact + SimHash near-duplicate detection 
import { codeIntelligencePlugin } from '@renseiai/agentfactory-code-intelligence'

// Register with orchestrator
const orchestrator = createOrchestrator({
  toolPlugins: [codeIntelligencePlugin],
})

Supported languages

  • TypeScript / JavaScript
  • Python
  • Go
  • Rust

License

MIT