Semantic Grep

Offline local semantic code search using embeddings to find and index code by meaning with llama.cpp, ONNX, or Ollama backends.

semgrepll - Local Semantic Code Search

Description

Local semantic grep using embeddings - 100% offline capable with llama.cpp, ONNX, and Ollama backends. Use when searching code semantically, finding code by meaning, or indexing projects for AI-powered search.

Triggers

• "search code semantically"
• "find code by meaning"
• "semantic grep"
• "index project for search"
• Commands: semgrep index, semgrep search

Actions

• Index projects: semgrep index /path/to/project
• Search: semgrep search "query"
• List: semgrep ls
• Remove: semgrep rm <project>

Environment Variables (optional)

Variable	Default	Description
EMBED_BACKEND	auto	Backend: llama, onnx, ollama
LLM_MODEL_PATH	-	Path to GGUF model (llama.cpp)
ONNX_MODEL_PATH	auto	Path to ONNX model
SEMGREP_BACKEND	auto	Storage: sqlite, lance
EMBED_MODEL	mxbai-embed-large-v1	Embedding model

Installation

bash

pip install semgrepll
# Optional: ONNX support
pip install semgrepll[onnx]

Examples

• "Index my project for semantic search" → runs semgrep index ./project
• "Find authentication code" → runs semgrep search "authentication"
• "Search for payment processing" → runs semgrep search "payment processing"

Requirements

• Python 3.10+
• One of: llama-cpp-python, onnxruntime, or Ollama running locally

Notes

• 100% offline - no external API calls
• Auto-detects fastest available backend
• Embeddings cached for fast re-indexing