soma-search 0.1.1

Local-first semantic code search

soma-search

Local-first semantic code search. Index a repository once, then search it with natural-language queries — fully offline after a one-time model download. No API keys, no cloud, no per-query cost.

Install

pip install soma-search

Platform support

Heads up (0.1.1): the only prebuilt wheel right now is for macOS (Apple Silicon / arm64). On that platform pip install soma-search just works.

On Linux, Windows, and Intel macOS, pip falls back to building from the source distribution, which needs a Rust toolchain (rustup) and a C compiler (build-essential / Xcode Command Line Tools) installed. If you don't have those, the install will fail — for now, use cargo install soma-search-cli instead. Prebuilt wheels for all platforms are on the way.

Usage

import soma
soma.index("./my-repo")
for hit in soma.search("parse a config file", k=5, path="./my-repo"):
    print(hit["path"], hit["start_line"], hit["score"])

Or use the soma CLI installed with the wheel:

soma index ./my-repo
soma search "validate auth token" -k 10 -p ./my-repo
soma serve ./my-repo          # MCP server for coding agents (Claude Code, etc.)

First run downloads the embedding model (nomic-embed-text-v1.5, a few hundred MB) to the standard HuggingFace cache. Re-running soma index only re-embeds files that changed.

License

Apache-2.0