### Start ContextWeaver MCP Server

Source: https://context7.com/bayu605/contextweaver/llms.txt

Use the `mcp` command to start the Model Context Protocol server, which communicates via stdio for integration with AI assistants.

```bash
# Start MCP server (communicates via stdio)
contextweaver mcp
```

--------------------------------

### Initialize ContextWeaver Configuration

Source: https://context7.com/bayu605/contextweaver/llms.txt

Use the `init` command to create the configuration directory and default environment file. A short alias `cw` is also available.

```bash
# Initialize ContextWeaver configuration
contextweaver init

# Or use the short alias
cw init
```

--------------------------------

### Initialize and Use SearchService

Source: https://context7.com/bayu605/contextweaver/llms.txt

Initializes the SearchService with custom configurations for retrieval, reranking, and context expansion. Use this to build context packs for complex queries.

```typescript
import { SearchService } from '@hsingjui/contextweaver';
import { generateProjectId } from '@hsingjui/contextweaver';

// Initialize search service
const projectPath = '/path/to/project';
const projectId = generateProjectId(projectPath);

const searchService = new SearchService(projectId, projectPath, {
  // Optional: Override default search configuration
  vectorTopK: 200,      // Initial vector retrieval count
  fusedTopM: 50,        // Candidates after RRF fusion
  rerankTopN: 20,       // Results after reranking
  neighborHops: 2,      // Adjacent chunks to include
  breadcrumbExpandLimit: 3,  // Related chunks per breadcrumb
  importFilesPerSeed: 3,     // Cross-file imports to resolve
});

await searchService.init();

// Build context pack for a query
const contextPack = await searchService.buildContextPack(
  'How does the authentication middleware validate JWT tokens?'
);

console.log(contextPack);
```

--------------------------------

### Index a Codebase with ContextWeaver

Source: https://context7.com/bayu605/contextweaver/llms.txt

The `index` command scans a codebase and builds semantic indexes. You can index the current directory, a specific path, or force a re-index ignoring the cache.

```bash
# Index current directory
contextweaver index

# Index a specific path
contextweaver index /path/to/project

# Force re-index all files (ignores cache)
contextweaver index --force
contextweaver index -f /path/to/project
```

--------------------------------

### Configure Embedding and Reranker Settings

Source: https://context7.com/bayu605/contextweaver/llms.txt

Set environment variables in `~/.contextweaver/.env` to configure embedding and reranker APIs. This includes API keys, base URLs, models, concurrency, dimensions, and top N values for reranking. Optional custom ignore patterns can also be specified.

```bash
# Embedding API Configuration (Required)
EMBEDDINGS_API_KEY=your-api-key-here
EMBEDDINGS_BASE_URL=https://api.siliconflow.cn/v1/embeddings
EMBEDDINGS_MODEL=BAAI/bge-m3
EMBEDDINGS_MAX_CONCURRENCY=10
EMBEDDINGS_DIMENSIONS=1024

# Reranker Configuration (Required)
RERANK_API_KEY=your-api-key-here
RERANK_BASE_URL=https://api.siliconflow.cn/v1/rerank
RERANK_MODEL=BAAI/bge-reranker-v2-m3
RERANK_TOP_N=20

# Custom Ignore Patterns (Optional, comma-separated)
IGNORE_PATTERNS=.venv,node_modules,custom_folder
```

--------------------------------

### Manage Database Operations

Source: https://context7.com/bayu605/contextweaver/llms.txt

Handles project ID generation, SQLite database initialization, and batch file metadata operations.

```typescript
import {
  generateProjectId,
  initDb,
  closeDb,
  getAllFileMeta,
  batchUpsert,
  batchDelete
} from '@hsingjui/contextweaver';

// Generate unique project ID based on path and directory birthtime
const projectId = generateProjectId('/path/to/project');
// Returns: '8a3f2b1c9d' (MD5 hash prefix)

// Initialize SQLite database
const db = initDb(projectId);
// Creates ~/.contextweaver/{projectId}/index.db

// Get all indexed file metadata
const fileMeta = getAllFileMeta(db);
// Map<string, { mtime, hash, size, vectorIndexHash }>

// Batch insert/update files
batchUpsert(db, [
  {
    path: 'src/index.ts',
    hash: 'abc123...',
    mtime: Date.now(),
    size: 1024,
    content: 'export * from "./api";',
    language: 'typescript',
    vectorIndexHash: null
  }
]);

// Batch delete files
batchDelete(db, ['src/old-file.ts', 'src/deprecated.ts']);

// Close database connection
closeDb(db);
```

--------------------------------

### Search Codebase with ContextWeaver

Source: https://context7.com/bayu605/contextweaver/llms.txt

The `search` command performs semantic code retrieval using natural language queries. It supports filtering with technical terms, specifying a repository path, and disabling Zen mode for more cross-file context.

```bash
# Search with semantic query
contextweaver search --information-request "How is user authentication handled?"

# Search with technical terms for precise filtering
contextweaver search \
  --information-request "Logic for database connection pooling" \
  --technical-terms "PoolConfig,Connection,release"

# Search in a specific repository
contextweaver search \
  --repo-path /path/to/project \
  --information-request "How does the payment processing work?"

# Disable Zen mode (includes more cross-file context)
contextweaver search --no-zen --information-request "Error handling patterns"
```

--------------------------------

### MCP Server Configuration for ContextWeaver

Source: https://context7.com/bayu605/contextweaver/llms.txt

Configuration for the MCP server within an AI assistant's settings. Specifies the command and arguments to run the ContextWeaver MCP service.

```json
{
  "mcpServers": {
    "contextweaver": {
      "command": "npx",
      "args": ["@hsingjui/contextweaver", "mcp"]
    }
  }
}
```

--------------------------------

### Handle Codebase Retrieval for AI Assistants

Source: https://context7.com/bayu605/contextweaver/llms.txt

Integrates with MCP servers to provide a codebase retrieval tool for AI assistants. Call the retrieval handler directly with repository path, information request, and technical terms.

```typescript
import { handleCodebaseRetrieval } from '@hsingjui/contextweaver';

// Call the retrieval handler directly
const response = await handleCodebaseRetrieval({
  repo_path: '/path/to/project',
  information_request: 'How is user session management implemented?',
  technical_terms: ['SessionStore', 'createSession', 'validateSession']
});

console.log(response.content[0].text);
```

--------------------------------

### Initialize and Use SemanticSplitter with AST

Source: https://context7.com/bayu605/contextweaver/llms.txt

Initializes the SemanticSplitter using Tree-sitter for AST parsing to create semantically meaningful code chunks. Requires a parser and language definition.

```typescript
import { SemanticSplitter } from '@hsingjui/contextweaver';
import Parser from '@keqingmoe/tree-sitter';
import TypeScript from 'tree-sitter-typescript';

// Initialize parser
const parser = new Parser();
parser.setLanguage(TypeScript.typescript);

// Create splitter with custom configuration
const splitter = new SemanticSplitter({
  maxChunkSize: 2500,    // Max tokens per chunk
  minChunkSize: 100,     // Min tokens per chunk
  chunkOverlap: 200,     // Overlap for context continuity
  maxRawChars: 10000,    // Physical character limit
});

// Split code into semantic chunks
const code = "\nexport class UserService {\n  private db: Database;\n\n  constructor(db: Database) {\n    this.db = db;\n  }\n\n  async findById(id: string): Promise<User | null> {\n    return this.db.users.findUnique({ where: { id } });\n  }\n\n  async createUser(data: CreateUserDto): Promise<User> {\n    return this.db.users.create({ data });\n  }\n}
";

const tree = parser.parse(code);
const chunks = splitter.split(tree, code, 'src/services/user.ts', 'typescript');

console.log(chunks);
```

--------------------------------

### Scan Codebase with ContextWeaver

Source: https://context7.com/bayu605/contextweaver/llms.txt

The `scan` function crawls a codebase, processes files, generates embeddings, and builds the search index. It can be used with basic options or advanced options including a progress callback.

```typescript
import { scan, type ScanStats, type ScanOptions } from '@hsingjui/contextweaver';

// Basic scan
const stats: ScanStats = await scan('/path/to/project');
console.log(stats);
// {
//   totalFiles: 150,
//   added: 120,
//   modified: 10,
//   unchanged: 15,
//   deleted: 5,
//   skipped: 0,
//   errors: 0,
//   vectorIndex: { indexed: 130, deleted: 5, errors: 0 }
// }

// Scan with options and progress callback
const options: ScanOptions = {
  force: false,        // Set true to re-index all files
  vectorIndex: true,   // Enable vector indexing (default: true)
  onProgress: (current, total, message) => {
    console.log(`Progress: ${current}/${total} - ${message}`);
  }
};

const result = await scan('/path/to/project', options);
```

--------------------------------

### TypeScript Configuration Functions

Source: https://context7.com/bayu605/contextweaver/llms.txt

Import and use functions like `getEmbeddingConfig`, `getRerankerConfig`, `checkEmbeddingEnv`, `checkRerankerEnv`, and `getExcludePatterns` to manage ContextWeaver's configuration in TypeScript projects. These functions help validate environment variables and retrieve configuration objects.

```typescript
import {
  getEmbeddingConfig,
  getRerankerConfig,
  checkEmbeddingEnv,
  checkRerankerEnv,
  getExcludePatterns
} from '@hsingjui/contextweaver';

// Check if environment is properly configured
const embeddingCheck = checkEmbeddingEnv();
if (!embeddingCheck.isValid) {
  console.log('Missing variables:', embeddingCheck.missingVars);
}

// Get embedding configuration
const embeddingConfig = getEmbeddingConfig();
console.log(embeddingConfig);
// {
//   apiKey: 'your-api-key',
//   baseUrl: 'https://api.siliconflow.cn/v1/embeddings',
//   model: 'BAAI/bge-m3',
//   maxConcurrency: 10,
//   dimensions: 1024
// }

// Get reranker configuration
const rerankerConfig = getRerankerConfig();
console.log(rerankerConfig);
// {
//   apiKey: 'your-api-key',
//   baseUrl: 'https://api.siliconflow.cn/v1/rerank',
//   model: 'BAAI/bge-reranker-v2-m3',
//   topN: 20
// }

// Get exclude patterns for file scanning
const patterns = getExcludePatterns();
// ['node_modules', '.venv', 'dist', '*.min.js', ...]
```

--------------------------------

### Pack Context with ContextPacker

Source: https://context7.com/bayu605/contextweaver/llms.txt

Merges overlapping code chunks and enforces token limits to prepare data for LLM consumption.

```typescript
import { ContextPacker } from '@hsingjui/contextweaver';
import type { SearchConfig, ScoredChunk } from '@hsingjui/contextweaver';

const config: Partial<SearchConfig> = {
  maxSegmentsPerFile: 5,     // Max segments per file in output
  maxTotalChars: 50000,      // Total character budget
};

const packer = new ContextPacker(projectId, config as SearchConfig);

// Pack scored chunks into final context
const chunks: ScoredChunk[] = [
  { filePath: 'src/auth.ts', chunkIndex: 2, score: 0.95, record: { raw_start: 100, raw_end: 500, ... } },
  { filePath: 'src/auth.ts', chunkIndex: 3, score: 0.88, record: { raw_start: 450, raw_end: 800, ... } },
  { filePath: 'src/utils.ts', chunkIndex: 1, score: 0.75, record: { raw_start: 0, raw_end: 300, ... } },
];

const files = await packer.pack(chunks);

console.log(files);
// [
//   {
//     filePath: 'src/auth.ts',
//     segments: [
//       {
//         filePath: 'src/auth.ts',
//         rawStart: 100,
//         rawEnd: 800,      // Merged overlapping chunks
//         startLine: 5,
//         endLine: 35,
//         score: 0.95,
//         breadcrumb: 'src/auth.ts > validateToken',
//         text: '...'
//       }
//     ]
//   },
//   {
//     filePath: 'src/utils.ts',
//     segments: [...]
//   }
// ]
```

--------------------------------

### Expand Graph Context with GraphExpander

Source: https://context7.com/bayu605/contextweaver/llms.txt

Uses neighbor expansion, breadcrumb matching, and import resolution to enrich seed chunks with related code context.

```typescript
import { getGraphExpander } from '@hsingjui/contextweaver';
import type { ScoredChunk, SearchConfig } from '@hsingjui/contextweaver';

const config: Partial<SearchConfig> = {
  neighborHops: 2,           // Include ±2 adjacent chunks
  breadcrumbExpandLimit: 3,  // Max chunks per breadcrumb prefix
  importFilesPerSeed: 3,     // Max import files to resolve per seed
  chunksPerImportFile: 2,    // Chunks per imported file
  decayNeighbor: 0.9,        // Score decay for neighbors
  decayBreadcrumb: 0.8,      // Score decay for breadcrumb matches
  decayImport: 0.7,          // Score decay for imports
};

const expander = await getGraphExpander(projectId, config as SearchConfig);

// Expand seed chunks with related context
const seeds: ScoredChunk[] = [
  { filePath: 'src/api/handler.ts', chunkIndex: 5, score: 0.95, source: 'vector', record: {...} }
];

const queryTokens = new Set(['handler', 'request', 'response']);
const { chunks, stats } = await expander.expand(seeds, queryTokens);

console.log(stats);
// {
//   neighborCount: 4,      // Adjacent chunks added
//   breadcrumbCount: 2,    // Related function chunks
//   importCount: 3,        // Cross-file imports resolved
//   importDepth1Count: 1   // Barrel file re-exports followed
// }
```

--------------------------------

### Split Plain Text with SemanticSplitter

Source: https://context7.com/bayu605/contextweaver/llms.txt

Uses the SemanticSplitter for plain text splitting when AST parsing is not applicable. Specify the text, file path, and language.

```typescript
// For non-AST languages, use plain text splitting
const plainChunks = splitter.splitPlainText(
  'some plain text content...', 
  'README.md',
  'markdown'
);
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.