Sidecar AI

A native, privacy-first AI chat companion for Obsidian powered directly by your authenticated local CLI tools: Google Antigravity (agy), GitHub Copilot (copilot), Pi Coding Agent (pi), or any Custom CLI with zero API keys, proxies, or per-token charges.

✨ Why Sidecar AI?

Most AI plugins for Obsidian require you to manage paid third-party API tokens, configure proxies, or paste secret keys into plugin settings.

Sidecar AI bridges your vault directly to the command-line agents already running on your machine. It utilizes your existing subscriptions (Google AI subscriptions, GitHub Copilot, Pi Coding Agent, or your own custom agent scripts) in a native sidebar designed to match Obsidian's interface.

⚡ Quick Start

1. Install & Enable: Click Install and then Enable in Obsidian Settings → Community Plugins.
2. Ensure CLI is Ready: Make sure at least one supported tool is installed and logged in:
   • Google Antigravity: run agy in your terminal.
   • GitHub Copilot: run copilot login in your terminal.
   • Pi Coding Agent: install via npm install -g @earendil-works/pi-coding-agent and run pi.
   • Custom CLI: have your custom binary (e.g. super-ai) available in your terminal or PATH.
3. Open Sidecar AI:
   • Click the Sidecar AI bot icon in the left ribbon, or
   • Press Ctrl/Cmd + P and select Sidecar AI: Open chat sidebar.
4. Chat with Context: Ask questions, summarize notes, or select text in your active note to query the AI assistant.

🔌 Supported Providers

🚀 Key Features

• 🔐 Zero API Keys Needed: Interacts directly with authenticated local CLI processes. No keys or billing details are ever stored in Obsidian.
• 💬 Multi-Turn Conversation Memory: Subsequent messages within a chat session retain full dialogue history and note context across all providers.
• 🛠️ Custom CLI & Agent Support: Plug in any custom coding agent or local CLI with configurable arguments templates (e.g. tell {prompt}).
• 🌐 Cross-Platform & WSL Support: Native execution on Windows, macOS, Linux, and Windows Subsystem for Linux (wsl).
• 🎯 Unified Model & Reasoning Switcher: A single fuzzy-search modal accessible from the chat header, the bottom status bar widget, and plugin settings to switch providers, models, and reasoning effort levels on the fly.
• 📄 Active Note & Selection Tracking: Automatically references your currently focused note path (The context: "<path>":) and text selection as context.
• 🏷️ Clear Attribution: Every assistant response clearly displays the provider, model, and reasoning effort used.
• 🔘 Quick Action Bar: One-click prompt chips to Summarize, Polish writing, Extract tasks, and Explain concepts from your active note.
• 🔄 Session Management: Dedicated New session button to clear conversation memory and restart context on demand.
• 📋 Code Snippet Actions: One-click buttons to Copy code blocks or Insert into note at cursor.
• 🎨 Obsidian Native Design: Strictly styled with Obsidian theme CSS variables for seamless Dark and Light mode support.

⚙️ Settings Overview

🔒 Security & System Permissions

Sidecar AI uses a privacy-first, zero-telemetry architecture:

• Shell Execution (child_process.spawn):
  • The plugin spawns local CLI processes solely to communicate with your user-installed and authenticated AI CLI tools (agy, copilot, pi, or custom agent).
  • It executes exclusively the command configured in Settings (prefixed with wsl if enabled), passing user prompts via standard CLI arguments and reading streaming output via stdout. No background scripts, remote downloads, or arbitrary commands are executed.
• Clipboard Access (navigator.clipboard.writeText):
  • Clipboard access is strictly write-only.
  • It is triggered solely when you explicitly click the "Copy" button on code blocks in the chat UI.
  • The plugin never reads, inspects, or monitors system clipboard contents.

⌨️ Command Palette Shortcuts

📚 Advanced Setup & Manual Installation

For instructions on building from source, manual installation from GitHub releases, or installing via Obsidian BRAT for beta testing, see the Development & Manual Installation Guide.

📄 License

Licensed under the MIT License.

🤖 Development & Transparency

• This extension was built using Google Gemini AI.
• Each change, implementation detail, and line of code was reviewed and verified by the author.