semgrepll 1.3.1

Local semantic code search - multi-backend (llama.cpp, HF API, ONNX, Ollama)

semgrepll - Local Semantic Code Search

Local semantic grep using embeddings - 100% offline capable.

Features

• Multi-backend support: llama.cpp, ONNX, Ollama
• Auto-detection: Automatically picks fastest available local backend
• Hybrid storage: SQLite (small projects) or LanceDB (large)
• 100% offline: No external connections
• Embedding caching: Fast re-indexing
• Semantic search: Find code by meaning, not just keywords

Backends (priority order)

Rank	Backend	Speed	Notes
1	llama.cpp	~1s	Fastest - requires GGUF model
2	ONNX	~3s	Local - uses HF model files
3	Ollama	~6s	Local server fallback

Installation

pip install semgrepll

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

Usage

# Index a project (auto-detects backend)
semgrep index /path/to/project

# Search semantically
semgrep search "authentication logic"

# List indexed projects
semgrep ls

Environment Variables

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

Benchmark

Backend	Speed
llama.cpp	~1s
ONNX	~3s
Ollama	~6s

License

MIT License - see LICENSE for details.