semgrepll 1.2.0

Local semantic code search - hybrid SQLite/LanceDB, works offline with Ollama

semgrepll

Local semantic code search for AI agents and developers

Search your codebase using natural language — no API keys, no cloud, 100% offline.

What is semgrepll?

semgrepll (pronounced "sem-grep-ell") is a local semantic code search tool that lets you search your codebase using natural language queries. Unlike traditional grep that searches for exact text matches, semgrepll understands meaning — so you can ask questions like:

• "Where is authentication configured?"
• "How does the user login flow work?"
• "Find the payment processing logic"

Why semgrepll?

For Developers

• Offline & Private — Your code never leaves your machine
• No API Keys — Works without OpenAI, Anthropic, or any cloud service
• Fast — Local Ollama embeddings, SQLite/LanceDB storage
• Universal — Works with any programming language

For AI Agents

• Understand Codebases — Semantic search helps agents navigate unfamiliar code
• Reduce Token Usage — Instead of reading entire files, find exact locations
• Faster Context — Get relevant code sections in milliseconds

Comparison

Tool	Type	Requires API	Offline	Best For
semgrepll	Semantic	❌ No	✅ Yes	Local AI dev
GitHub Copilot	Semantic	✅ Yes	❌ No	Cloud IDEs
ripgrep (rg)	Exact	❌ No	✅ Yes	Known patterns
Sourcegraph	Semantic	✅ Yes	❌ No	Enterprise

Installation

# Basic (SQLite backend - works out of the box)
pip install semgrepll

# With LanceDB (recommended for large projects)
pip install semgrepll[lance]

Requirements

• Python 3.10+
• Ollama running locally with mxbai-embed-large model

# Install Ollama and the embedding model
ollama pull mxbai-embed-large

Quick Start

# 1. Index your project (one-time)
semgrep index /path/to/your/project

# 2. Search semantically
semgrep search "how does authentication work"

# 3. List indexed projects
semgrep ls

# 4. Remove a project
semgrep rm project-name

Usage

CLI Commands

semgrep index <path>           # Index a project for search
semgrep search <query>         # Search indexed code
semgrep ls                     # List all indexed projects
semgrep rm <project>           # Remove a project index

Options

semgrep search "query"         # Search all indexed projects
semgrep search "query" -p myproject  # Search specific project
semgrep search "query" -e "pattern"  # Fallback to ripgrep

Configuration

Environment Variables

# Ollama endpoint (default: http://127.0.0.1:11434)
export OLLAMA_URL="http://localhost:11434/api/embeddings"

# Embedding model (default: mxbai-embed-large)
export EMBED_MODEL="mxbai-embed-large"

# Storage backend (auto | sqlite | lance)
# - auto: SQLite for small projects, LanceDB for large
# - sqlite: Force SQLite (no extra deps)
# - lance: Force LanceDB (needs lancedb installed)
export SEMGREP_BACKEND=auto

# Database path (default: ~/.semgrepll/db)
export SEMGREP_DB_PATH="/path/to/db"

When to Use Which Backend

Project Size	Recommended Backend	Why
Small (< 100 files)	SQLite	Zero deps, fast enough
Large (100+ files)	LanceDB	Better vector indexing
Mixed	auto	Automatic selection

How It Works

┌─────────────┐     ┌──────────────┐     ┌─────────────┐
│   Your      │────▶│   semgrepll   │────▶│   Ollama    │
│   Query     │     │  (embed)      │     │ (mxbai)     │
└─────────────┘     └──────────────┘     └─────────────┘
                                                  │
                     ┌──────────────┐            │
                     │   SQLite or   │◀───────────┘
                     │   LanceDB     │
                     │  (similarity) │
                     └──────────────┘

1. Index — Your code files are chunked and embedded using Ollama
2. Search — Your query is embedded, then compared against indexed code
3. Results — Most similar code sections returned with relevance scores

Use Cases

Developer Onboarding

# New to the codebase? Ask questions!
semgrep search "how do I add a new API route"
semgrep search "where is error handling"

AI Agent Integration

# In your AI agent
subprocess.run(["semgrep", "search", "-p", "myproject", "auth configuration"])
# Returns: file paths + code snippets + relevance scores

Code Review

# Find all places that touch payments
semgrep search "payment processing"

Example Output

🔍 Searching: how does authentication work

📄 auth.ts (score: 0.85)
   // Authentication module
   export class AuthService {
     async signIn(email: string, password: string) {
       return this.client.auth.signInWithPassword({...});
     }
   }

📄 middleware.ts (score: 0.72)
   export function authMiddleware(request: NextRequest) {
     const token = request.headers.get('authorization');
     ...
   }

Contributing

# Clone and develop
git clone https://github.com/rizperdana/semgrepll
cd semgrepll
pip install -e ".[all]"
pip install pytest black mypy

# Run tests
pytest

# Format
black semgrepll/

License

MIT License — see LICENSE for details.

Related

• Ollama — Run LLMs locally
• LanceDB — Vector database
• ripgrep — Fast line-oriented search

---

TL;DR: pip install semgrepll → semgrep index ./src → semgrep search "how does X work"