SnCode

A desktop AI coding agent. Open a project, chat with the agent, and let it read, write, and run code for you.

---

What is SnCode?

SnCode is a local-first desktop application that gives you an AI coding agent with full access to your project files and terminal. It works like Claude Code or Cursor's agent mode, but as a standalone Electron app.

You open a folder, start a conversation, and the agent uses built-in tools to explore, read, edit, and run commands in your codebase — all streamed in real-time.

Features

Agent & Tools

• 9 built-in tools: list_files, read_file, write_file, edit_file, glob, grep, run_command, load_skill, spawn_task
• Native streaming with Anthropic (Claude) and OpenAI (Codex) providers
• Sub-agent system — delegate to general (full access) or explore (read-only) sub-agents
• Parallel sub-agent execution with configurable concurrency
• Extended thinking / reasoning effort control (Anthropic up to High, Codex up to X-High)

Workspace

• Multi-project workspace with multiple threads per project
• AI-generated thread titles using the smallest available model
• File tree panel in the sidebar with click-to-preview
• Git integration: branch switching, commit, pull, push, stash, diff viewer
• Git repo initialization when no repo is detected
• Keyboard shortcuts (Ctrl+N, Ctrl+W, Ctrl+F, Ctrl+B, Ctrl+,)

UI

• Dark and light themes with CSS custom properties
• Syntax-highlighted code blocks (highlight.js, 16+ languages)
• Inline search with match highlighting and navigation
• Right sidebar for file preview, git diffs, and sub-agent details
• IDE-style diff viewer with line numbers, hunk headers, and status badges
• Image support (paste, drag-and-drop, file upload)
• Markdown rendering with GFM support
• Token usage display per response
• Todo/task checklist above the input area

Skills System

• Discover, load, and enable SKILL.md files from multiple sources
• Agent can dynamically load skills via the load_skill tool
• Per-project skill configuration

Security

• Credentials stored in OS keychain via keytar
• OAuth flows for Anthropic and OpenAI
• Sandboxed renderer with context isolation
• Tool operations scoped to project root
• Path traversal protection on file reads

Installation

From Source

git clone https://github.com/Snowy7/sncode.git
cd sncode
npm install
npm run dev

Build & Package

# Production build
npm run build

# Package for distribution (Windows / macOS / Linux)
npm run package

Installers and portables are output to release/.

Requirements

• Node.js 20+
• Git (for git integration features)
• An API key or OAuth token for Anthropic and/or OpenAI

Tech Stack

Layer	Technology
Framework	Electron 37
Frontend	React 19, TypeScript 5.9
Styling	Tailwind CSS v4
AI SDKs	@anthropic-ai/sdk, openai
Credentials	keytar (OS keychain)
Validation	Zod
Code Highlighting	highlight.js
Markdown	react-markdown + remark-gfm
Bundler	Vite 7
Testing	Vitest
Packaging	electron-builder

Project Structure

src/
  main/           # Electron main process
    main.ts        - Window, IPC handlers, agent orchestration
    agent.ts       - AI agent loop with streaming & tool calling
    store.ts       - JSON persistence
    project-tools.ts - File/command tools
    mcp.ts         - MCP client (in progress)
    skills.ts      - Skills discovery & loading
    oauth.ts       - OAuth flows
    credentials.ts - Keychain access
    preload.ts     - Context bridge
  renderer/       # React frontend
    App.tsx        - Main UI
    SettingsModal.tsx - Settings panel
    styles.css     - Theme variables & base styles
  shared/         # Shared between main & renderer
    types.ts       - All interfaces & types
    schema.ts      - Zod validation schemas
    models.ts      - Model catalog & helpers

Verification

npm run lint        # ESLint
npm run typecheck   # TypeScript (both configs)
npm run test        # Vitest (33 tests)
npm run build       # Vite + tsc production build

Contributing

See CONTRIBUTING.md for guidelines.

License

MIT