실험 – Claude Code의 컨텍스트 스터핑을 의미론적 grep으로 대체

Semantic search for your codebase. Parses every tracked file with tree-sitter, generates vector embeddings per chunk, and stores them on a dedicated orphan Git branch that mirrors your source tree — so the whole team can share embeddings without re-indexing. main branch semantic branch (orphan) ────────────────── ────────────────────────────── src/main.rs → src/main.rs ← [{start_line, end_line, text, embedding}, ...] src/db.rs → src/db.rs ← [{...}, ...] src/chunking/mod.rs → src/chunking/mod.rs git-semantic index parses all tracked files with tree-sitter, embeds each chunk, and commits the mirrored JSON files to thesemantic orphan branch. On subsequent runs it only re-embeds files that changed since the last index (incremental)git push origin semantic shares the embeddings with the team- Contributors run git fetch origin semantic +git-semantic hydrate to populate their local SQLite search index — no re-embedding needed git-semantic grep runs KNN vector similarity search against the local index Indexing only needs to happen once — whoever runs it pushes the semantic branch and the whole team benefits. Nobody else needs an API key or has to re-embed anything. You can run indexing manually from any machine, or automate it in your CI/CD pipeline so embeddings stay fresh after every merge. # Anyone with an API key runs this once (or after significant changes) git-semantic index git push origin semantic # Everyone else git fetch origin semantic git-semantic hydrate git-semantic grep "..." Add .github/workflows/semantic-index.yml to your repository and indexing happens automatically on every merge to main: name: Semantic Index on: push: branches: [main] jobs: index: runs-on: ubuntu-latest steps: - uses: actions/checkout@v4 with: fetch-depth: 0 token: ${{ secrets.GITHUB_TOKEN }} - name: Install git-semantic run: cargo install git-semantic - name: Index codebase env: OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }} run: git-semantic index - name: Push semantic branch run: | git config user.name "github-actions[bot]" git config user.email "github-actions[bot]@users.noreply.github.com" git push origin semantic - Rust 1.65 or higher - Git 2.0 or higher cargo install git-semantic git clone https://github.com/ccherrad/git-semantic.git cd git-semantic cargo install --path . Parses and embeds files, then commits the result to the semantic orphan branch. git-semantic index - First run: full index of all tracked files, writes .indexed-at with the current HEAD SHA - Subsequent runs: incremental — diffs against the last indexed SHA, re-embeds only added, modified, renamed, or deleted files - Respects .gitignore (usesgit ls-files ) - Skips binary files - Files with unrecognized extensions are stored as a single chunk - Creates the semantic branch automatically on first run Reads the semantic branch and populates the local .git/semantic.db search index. git-semantic hydrate Attempts to fetch origin/semantic first, then falls back to the local branch. Search code semantically using natural language. git-semantic grep "authentication logic" git-semantic grep "error handling" -n 5 Injects code search instructions into CLAUDE.md so coding agents automatically use git-semantic grep instead of git grep . git-semantic agentic-setup - Appends instructions to an existing CLAUDE.md , or creates one if it doesn't exist - Idempotent — safe to run multiple times - Works with Claude Code, Cursor, GitHub Copilot (via .cursor/rules or.github/copilot-instructions.md equivalents) Configure the embedding provider. git-semantic config --list git-semantic config gitsem.provider openai git-semantic config gitsem.provider onnx git-semantic config --get gitsem.provider git-semantic config --unset gitsem.onnx.modelPath export OPENAI_API_KEY="sk-..." git-semantic config gitsem.provider openai git-semantic config gitsem.provider onnx git-semantic config gitsem.onnx.modelPath /path/to/model.onnx Rust, Python, JavaScript, TypeScript, Java, C, C++, Go git-semantic/ ├── src/ │ ├── main.rs # CLI and command handlers │ ├── models.rs # CodeChunk data structure │ ├── db.rs # SQLite + sqlite-vec search index │ ├── embed.rs # Embedding generation │ ├── semantic_branch.rs # Orphan branch read/write via git worktree │ ├── embeddings/ # OpenAI and ONNX provider implementations │ └── chunking/ # tree-sitter parsing and language detection ├── Cargo.toml └── README.md cargo build --release cargo test MIT OR Apache-2.0