@grepai/cli

@grepai/cli

Semantic code search CLI. It indexes your codebase using vector embeddings to enable natural language queries that find relevant code by meaning, regardless of naming conventions.

Installation

bun add -g @grepai/cli

Usage

# Index your codebase (run whenever code changes)
grepai index

# Search using natural language
grepai search "how is user authentication handled?"

# Options: --topK (default 20)
grepai search "database config" --topK 50

Configuration

Create a .grepairc.json or grepai-config.json in your project root. Supports ${ENV_VAR} substitution.

Minimal Example (Turso)

{
  "$schema": "https://raw.githubusercontent.com/bismuth1991/grepai-ts/refs/heads/main/packages/core/src/grepai-config-schema.json",
  "storage": {
    "type": "turso",
    "url": "${TURSO_DB_URL}",
    "authToken": "${TURSO_DB_AUTH_TOKEN}"
  },
  "embedding": {
    "provider": "google",
    "model": "gemini-embedding-001",
    "apiKey": "${GEMINI_API_KEY}"
  },
  "include": ["src/**/*.ts"],
  "exclude": ["node_modules", "dist", "**/*.test.ts"]
}

Postgres Example

{
  "storage": {
    "type": "postgres",
    "connectionString": "postgres://user:pass@localhost:5432/grepai"
  },
  "embedding": {
    "provider": "google",
    "model": "gemini-embedding-001",
    "apiKey": "${GEMINI_API_KEY}",
    "dimensions": 1536
  }
}

Local LanceDB Example

{
  "storage": {
    "type": "lancedb"
  },
  "embedding": {
    "provider": "google",
    "model": "gemini-embedding-001",
    "apiKey": "${GEMINI_API_KEY}",
    "dimensions": 1536
  }
}

Advanced Configuration

Full list of options:

{
  "embedding": {
    "provider": "google",
    "model": "gemini-embedding-001",
    "apiKey": "${GEMINI_API_KEY}",
    "dimensions": 3072, // 768, 1536, or 3072
    "targetChunkSize": 256, // Preferred tokens per chunk
    "maxChunkSize": 1024, // Max tokens per chunk
    "tokenizer": "gemini-embedding-001" // or "simple"
  }
}

Read https://github.com/bismuth1991/grepai-ts/blob/main/packages/core/src/domain/config.ts for full configuration details.

How it works

• AST Parsing: GrepAI parses your code (TypeScript/TSX) into an Abstract Syntax Tree.
• Smart Chunking: It splits code into meaningful chunks, preserving scope context (class names, function signatures) even for deep nested code.
• Embedding: Chunks are embedded using Google's Gemini models.
• Vector Search: Queries are embedded and compared against code chunks using cosine similarity.

Links

Repository | Issues | License