snipcontext 0.3.0

AI-powered code snippet & context manager with local semantic search

SnipContext

AI-powered code snippet & context manager.

Save, search, tag, and instantly inject your best boilerplate, patterns, and context into any LLM (Claude, Cursor, Grok, Windsurf, etc.).

Local-first — Open source — Built for humans + AI agents

🎧 Stop Feeding Your AI Clipboard Garbage — Why SnipContext exists.

Searching, tagging, and exporting code snippets — all from the terminal. Watch the animated demo (GIF)

---

Why SnipContext?

• Stop rewriting the same auth flows, component patterns, or utility functions
• Stop feeding LLMs messy or outdated code from your clipboard history
• Build your personal/team "second brain" of high-quality, reusable code
• Semantic search finds code by meaning, not just keywords
• LLM-optimized exports format your snippets for maximum comprehension

---

Key Features

Feature	Status	Description
Rich snippet saving with tags, metadata, and versioning	✅	Full CRUD with soft-delete and encryption
Semantic search with local embeddings	✅	sentence-transformers + FAISS, runs offline
Hybrid search — semantic + keyword fusion	✅	Configurable weights, TF-IDF + embeddings
LLM-optimized export providers	✅	Claude XML, Cursor, OpenAI, Generic Markdown
Auto-tagging via embeddings	✅	Suggests tags based on similar snippets
Similarity-based deduplication	✅	Warns when adding near-duplicate snippets
Encryption at rest	✅	Fernet (AES-128) with PBKDF2 key derivation
File watchdog / real-time indexing	✅	Auto-reindex on file changes
Plugin system	✅	Entry points for providers and exporters
CLI + Python library	✅	Use from terminal or import as a module
Git-friendly local-first storage	✅	One JSON file per snippet, easy to version

Supported LLM Providers

Provider	Format	Best For
Generic	Markdown	Universal compatibility
Claude	XML documents	Anthropic Claude
Cursor	File-style headers	Cursor IDE
OpenAI	Delineated sections	ChatGPT / GPT-4

---

Quick Start

Installation

# From PyPI with uv (recommended — faster installs, better dependency resolution)
uv tool install snipcontext

# From PyPI with pip
pip install snipcontext

# From source (after cloning)
cd snipcontext
uv sync                    # install all deps (including dev)
uv run sc --help           # run without activating venv

# Or with pip (traditional)
pip install -e ".[dev]"

💡 Why uv? This project uses `uv` for dependency management (`uv.lock` pinned). `uv sync` guarantees reproducible installs. `pip install` works but may resolve dependencies differently.

Or install directly from GitHub

pip install git+https://github.com/billybox1926-jpg/snipcontext.git

> **📦 Dependency Footprint:** SnipContext's core (add, list, edit, delete, keyword search, export) has no heavy dependencies. Optional features are split into extras:
> - `pip install snipcontext[semantic]` — semantic search with sentence-transformers + FAISS (~500MB, requires Rust toolchain on ARM)
> - `pip install snipcontext[encryption]` — encryption at rest with Fernet/AES-128 (requires Rust toolchain on ARM)
> - `pip install snipcontext[tui]` — interactive terminal UI
> - `pip install snipcontext[all]` — all optional features
>
> **Lighter embedding model:** The default model is `all-MiniLM-L6-v2` (~80MB). For a lighter alternative, set `SNIPCONTEXT_EMBED_MODEL_NAME=all-MiniLM-L4-v2` (~30MB) or `SNIPCONTEXT_EMBED_MODEL_NAME=paraphrase-MiniLM-L3-v2` (~20MB) before searching.
>
> **Skip semantic at runtime:** Even with `pip install snipcontext[semantic]`, use `--no-semantic` to force keyword-only search for faster results:
> ```bash
> snipcontext search "hello world" --no-semantic
> ```
>
> **ARM / Android / Termux:** The `semantic` and `encryption` extras require Rust to build native wheels. On platforms without pre-built wheels (ARM64, Android/Termux), install the core package only and use keyword search + export features. Semantic search and encryption gracefully degrade with clear error messages when their dependencies are missing.

> **Windows Users:** The short alias `sc` is shadowed by the Windows built-in `sc.exe` (Service Control). Three workarounds are available:
>
> 1. **Full command name** — always works after installation:
>    ```powershell
>    snipcontext add "print('hello')" --title "Hello" --tag python
>    ```
> 2. **Wrapper script** — shipped automatically with `pip install`; adds `snipcontext.cmd` to your Scripts directory:
>    ```powershell
>    snipcontext.cmd search "hello world"
>    ```
> 3. **Shell alias** — for quick access in the current session:
>    ```cmd
>    doskey snip=python -m snipcontext $*
>    ```

### Standalone Binary

Two options for running without a Python environment:

**Option 1 — `uv tool` (recommended, lightweight):**

```bash
# Core features only (keyword search, export)
uv tool install snipcontext

# All features (semantic search, encryption, TUI, web)
uv tool install "snipcontext[all]"

# Use directly — uv manages the venv invisibly
snipcontext add "print('hello')" --title "Hello"

Option 2 — Pre-built binary (no Python needed):

Download from the latest GitHub Release. Two variants are available for each platform:

Variant	Includes	Size (approx.)
snipcontext-<platform>	Everything (semantic, encryption, TUI, web)	~200MB
snipcontext-<platform>-minimal	Core only (keyword search, export)	~80MB

# Linux / macOS
chmod +x snipcontext
./snipcontext search "hello world"

# Windows
snipcontext.exe search "hello world"

Build from source:

# Using Make
make build-binary           # full build
make build-binary-minimal   # core-only build

# Using PyInstaller directly
pip install pyinstaller
pyinstaller snipcontext.spec
# Output: dist/snipcontext (or dist/snipcontext.exe)

Security Considerations

• Encryption at rest: Uses Fernet (AES-128-CBC with HMAC) with PBKDF2 key derivation (100k iterations). Passphrase is read from SNIPCONTEXT_ENCRYPTION_PASSPHRASE env var — never pass it on the command line (shell history leak).
• No default passphrase: If encryption is enabled but SNIPCONTEXT_ENCRYPTION_PASSPHRASE is not set, the tool raises an error rather than falling back to a known default. This prevents a false sense of security.
• stdin for sensitive content: Use sc add --file secret.py or pipe via stdin (cat secret.py | sc add --file) to avoid shell history leaks with --encrypt.
• Salt: Auto-generated on first use and persisted to the config file. Back up your config file to avoid losing access to encrypted snippets.
• No network calls: All processing is local. No data leaves your machine.

# Windows: use the full command name or the .cmd wrapper
snipcontext add "print('hello')" --title "Hello" --tag python
snipcontext search "hello world"
snipcontext list
snipcontext stats

# Or run via module
python -m snipcontext add "print('hello')" --title "Hello" --tag python

Verify Installation

snipcontext --help          # or: python -m snipcontext --help
snipcontext providers       # List available export providers

CLI Usage

# Add a snippet
snipcontext add "def authenticate(token):\n    return jwt.decode(token, SECRET)" \
  --title "JWT Authentication" \
  --desc "Decode and verify JWT tokens" \
  --lang python \
  --tag auth --tag jwt --tag security

# Search semantically
snipcontext search "how to validate auth tokens"

# Search by tag
snipcontext search "auth" --mode tag

# Export for Claude
snipcontext search "authentication" --provider claude --output context.xml

# List all snippets
snipcontext list

# Show stats
snipcontext stats

# Delete a snippet
snipcontext delete <snippet-id>

# Edit a snippet
snipcontext edit <snippet-id> --title "New Title" --add-tag python

# Rebuild search index
snipcontext build-index --force

# Watch for file changes and auto-reindex
snipcontext watch

# Run the demo
snipcontext demo

Library Usage

from snipcontext.core.models import Snippet, SnippetMetadata, Language
from snipcontext.core.storage import StorageEngine
from snipcontext.core.search import HybridSearch
from snipcontext.config.settings import get_config

# Initialize
config = get_config()
storage = StorageEngine(config)

# Create and save a snippet
snippet = Snippet(
    content="def memoize(fn):\n    cache = {}\n    ...",
    metadata=SnippetMetadata(
        title="Memoization Decorator",
        description="Cache function results",
        language=Language.PYTHON,
    ),
    tags=["python", "decorator", "performance"],
)
storage.save(snippet)

# Search with semantic understanding
searcher = HybridSearch(config)
searcher.index_snippets(storage.list_all())
results = searcher.search("cache function results decorator")

for r in results:
    print(f"{r.score:.3f} | {r.snippet.metadata.title}")

---

🔐 Encryption at Rest

SnipContext supports Fernet (AES-128) encryption for sensitive snippets. When enabled, snippet content is encrypted at rest using a key derived from a passphrase via PBKDF2 (100k iterations).

Enable Encryption

# Enable encryption (required)
export SNIPCONTEXT_ENCRYPT_ENABLED=true

# Set passphrase (used for key derivation)
export SNIPCONTEXT_ENCRYPTION_PASSPHRASE="your-secure-passphrase"

# Optional: persist salt to config (auto-generated if omitted)
export SNIPCONTEXT_ENCRYPT_KEY_SALT="base64-encoded-salt"

Encrypt Snippets

# Encrypt a new snippet
snipcontext add "api_key = 'sk-12345'" \
  --title "API Key" \
  --tag secret \
  --encrypt

# Mark as sensitive (auto-enables encryption)
snipcontext add "password = 'secret123'" \
  --title "DB Password" \
  --sensitive

Decrypt for Viewing/Editing

# Decrypt for viewing
snipcontext decrypt <snippet-id>

# Encrypt an existing snippet
snipcontext encrypt <snippet-id>

Note: When encrypted, the plaintext content is cleared from storage. The encrypted_content field stores the encrypted data. Use snipcontext decrypt <id> to restore plaintext for editing.

---

🔄 Index Rebuild & Resilience

SnipContext automatically detects and recovers from index corruption. The HybridSearch engine validates index integrity on load and rebuilds automatically when needed.

Manual Rebuild

# Build or rebuild the semantic search index
snipcontext build-index

# Force rebuild (useful after corruption, dependency changes, or mode switches)
snipcontext build-index --force

Auto-Recovery

The search engine automatically:

1. Validates index integrity on load (checks ID map lengths, matrix dimensions)
2. Cleans up corrupted files (deletes mismatched/corrupted index files)
3. Falls back gracefully — if semantic index unavailable, runs keyword-only search
4. Rebuilds on demand — index_snippets() auto-loads existing indices before rebuilding

Watchdog / Real-time Indexing

Run snipcontext watch to monitor the snippets directory and automatically reindex when files change:

snipcontext watch

Disable via config if you prefer manual rebuilds only:

export SNIPCONTEXT_WATCHDOG_ENABLED=false

---

Architecture

┌─────────────────────────────────────────────────┐
│                  CLI (Typer + Rich)              │
├──────────┬──────────┬──────────┬────────────────┤
│  add     │  search  │  export  │  edit/delete   │
│  list    │  stats   │  watch   │  demo          │
└────┬─────┴────┬─────┴────┬─────┴───────┬────────┘
     │          │          │             │
     ▼          ▼          ▼             ▼
┌─────────────────────────────────────────────────┐
│              Search Engine (HybridSearch)        │
│  ┌──────────────┐  ┌──────────────────────────┐ │
│  │   Semantic    │  │       Keyword            │ │
│  │  FAISS Index  │  │     TF-IDF (sklearn)     │ │
│  └──────────────┘  └──────────────────────────┘ │
├─────────────────────────────────────────────────┤
│              Storage Engine                      │
│         Git-friendly JSON per snippet            │
├─────────────────────────────────────────────────┤
│              Data Models (Pydantic v2)           │
│     Snippet / SnippetMetadata / Language         │
└─────────────────────────────────────────────────┘

See docs/ARCHITECTURE.md for detailed design documentation.

---

Configuration

SnipContext uses environment variables and a YAML config file:

# Use GPU for embeddings
export SNIPCONTEXT_EMBED_DEVICE="cuda"

# Change embedding model
export SNIPCONTEXT_EMBED_MODEL_NAME="all-mpnet-base-v2"

# Adjust search weights
export SNIPCONTEXT_SEARCH_SEMANTIC_WEIGHT="0.8"

# Enable auto-tagging
export SNIPCONTEXT_AUTO_TAG_ENABLED=true

# Enable deduplication
export SNIPCONTEXT_DEDUP_ENABLED=true
export SNIPCONTEXT_DEDUP_THRESHOLD="0.95"

Or edit ~/.config/SnipContext/snipcontext.yaml:

embedding:
  model_name: "all-MiniLM-L6-v2"
  device: "cpu"

search:
  default_mode: "hybrid"
  semantic_weight: 0.7
  keyword_weight: 0.3
  top_k: 10

auto_tag:
  enabled: true
  threshold: 0.75

dedup:
  enabled: true
  threshold: 0.95

---

Development

# Clone
git clone https://github.com/billybox1926-jpg/snipcontext.git
cd snipcontext

# Install dev dependencies
pip install -e ".[dev]"

# Run tests
pytest

# Run with coverage
pytest --cov=snipcontext

# Linting
ruff check .
mypy .

# Install pre-commit hooks
pre-commit install

---

Roadmap

• Core snippet CRUD with git-friendly storage
• Semantic + hybrid search with local embeddings
• LLM-optimized export providers (Claude, Cursor, OpenAI, Generic)
• Rich CLI with Typer
• Plugin system with entry points
• Python library distribution (PyPI)
• Auto-tagging and deduplication
• Encryption at rest
• File watchdog / real-time indexing
• Import from GitHub Gists
• Import from Git repositories
• Snippet templates and scaffolding
• Team sharing via git-sync
• VS Code extension

---

Project Structure

snipcontext/
├── src/snipcontext/          # Python package
│   ├── __init__.py
│   ├── __main__.py           # python -m snipcontext
│   ├── cli/
│   │   └── main.py           # Typer CLI commands
│   ├── config/
│   │   └── settings.py       # Pydantic Settings
│   ├── core/
│   │   ├── models.py         # Pydantic data models
│   │   ├── storage.py        # Git-friendly JSON storage
│   │   ├── search.py         # Semantic + hybrid search
│   │   ├── auto_tag.py       # Embedding-based auto-tagging
│   │   └── watcher.py        # File watchdog
│   ├── plugins/
│   │   └── base.py           # Plugin base + manager
│   └── providers/
│       ├── base.py           # Provider interface
│       ├── claude.py         # Anthropic Claude XML
│       ├── cursor.py         # Cursor IDE format
│       ├── openai.py         # OpenAI format
│       └── generic.py        # Universal Markdown
├── tests/                    # Test suite
├── docs/                     # Documentation
│   ├── ARCHITECTURE.md
│   ├── API.md
│   └── MAINTAINER.md
├── pyproject.toml
├── CHANGELOG.md
└── README.md

---

License & Contributing

MIT License — see LICENSE for details.

Contributions are welcome! Please read CONTRIBUTING.md and CODE_OF_CONDUCT.md first. New contributors should check out our Good First Issues.