AIBit-Discover Open Source Projects AIBit-Discover Open Source Projects

LobsterAI: Open‑Source AI Assistant Built on Electron

Discover LobsterAI, the all‑in‑one personal assistant powered by Claude and built with Electron. Packed with data analysis, PPT generation, email handling, and sandboxed execution, it runs on macOS, Windows, and Linux. Whether you’re a developer looking to hook your own skills or a power user seeking a free productivity boost, this guide walks you through installation, quick start, packaging, and expanding the skill set. Dive into the architecture, learn how to add new tools, and see how LobsterAI bridges your desktop with mobile IM platforms for on‑the‑go control.

LobsterAI: A Free, All‑in‑One AI Personal Assistant

LobsterAI (GitHub: netease-youdao/LobsterAI) is an open‑source project that turns your desktop into a 24/7 intelligence hub. Built with Electron, React, TypeScript and the Anthropic Claude Agent SDK, it can:

  • Generate documents – Word, PowerPoint, Excel, PDF, and video.
  • Search the web and pull facts in real time.
  • Automate browsers with Playwright.
  • Run tasks locally or inside a safe Alpine Linux sandbox.
  • Schedule recurring workflows via cron syntax.
  • Remotely trigger work from your phone through Telegram, Discord, DingTalk, Feishu or NetEase IM.

Below is a comprehensive walkthrough of LobsterAI’s architecture, how to install it, build installers, and extend its capabilities.

1. Overview

LobsterAI’s core is a Cowork system – a lightweight session manager that orchestrates Claude‑powered conversations, tool‑invocations and file manipulation.

Key architectural components:

Layer Technology Notes
Frontend React 18 + Redux Toolkit + Tailwind CSS UI, user settings, message stream
Backend Electron (Node ≥ 24) IPC, SQLite persistence, skill loader
AI Engine Claude Agent SDK Handles natural language requests
Execution Local Node or Alpine VM Sandbox for safety
IM Gateways Telegram, Discord, DingTalk, Feishu, NetEase IM Remote command channel

Why Electron?

Electron gives us a deterministic desktop environment. It couples the speed of Node.js for heavy IO (email, database) with Chromium for a polished UI. Context isolation, sandboxes and typed IPC keep the system secure.

2. Getting Started

Prerequisites

# Node 24 or any 2025 LTS is fine.
nvm install 24
# npm >= 10
npm --version

Clone and Install

git clone https://github.com/netease-youdao/LobsterAI.git
cd LobsterAI
npm install

Run in Development Mode

npm run electron:dev

The app will open at http://localhost:5175 and launch a Chromium window. You can edit React components and see hot re‑loads.

3. Building for Production

The repo uses electron‑builder. Build targets for macOS, Windows, and Linux are defined in electron-builder.json.

# TypeScript → bundle
npm run build
# Lint checks
npm run lint
# Build
npm run dist

The installers are placed in the release/ folder:

Platform File How to install
macOS (Intel) LobsterAI-0.1.21-darwin-x64.dmg Double‑click and drag to Applications
macOS (Apple‑Silicon) ...-arm64.dmg Same as above
Windows (.exe) ...-win32-x64.exe Run and follow the wizard
Linux (.AppImage) ...-linux-x86_64.AppImage chmod +x then execute

4. Extending LobsterAI – Adding a Skill

Each skill is a lightweight module that exposes a skill.json schema and an executable handler.

  1. Create a new folder under SKILLs/ (e.g., markdown/).
  2. Add skill.json with name, description, and entry points.
  3. Place a Node script or compiled binary for the skill.
  4. Enable it in skills.config.json.
  5. Restart the app.

The repo ships with 16 built‑in skills: web-search, pptx, remotion, playwright, imap-smtp-email, etc. Feel free to clone one and tweak for your own needs.

5. Remote Control via IM

LobsterAI listens on four IM platforms. Each integration requires a token/secret that you set in Settings → IM Gateways.

Once a token is present, the app exposes a bot that accepts a simple natural‑language command. For example:

Hey Lobster, generate a weekly summary PPT from the last 10 emails.

The bot authenticates, authorizes, then spawns a Cowork session and streams the result back to your phone.

6. Security Highlights

Feature Description
Context isolation No Node APIs in the renderer
Sandbox execution Optional Alpine VM for unsafe code
Permission gating User must approve file or web actions
SQLite on disk only No external database

7. Contributing

  1. Fork → clone.
  2. Create a topic branch (git checkout -b feat/new‑skill).
  3. Commit with a conventional message (feat: add markdown skill).
  4. Push and open a PR.

We welcome documentation, new skill templates, and bug‑fixes. Make sure to run npm run lint before submitting.

8. Usage Scenarios

Scenario How LobsterAI Helps
Data analytics Import an Excel file → ask AI to produce insights and a PPT chart
Email triage Auto‑classify inbox, summarize important threads
Video marketing Pass a script → generate a Remotion video in seconds
Daily news digests Schedule a recurring task that scrapes tech sites and emails you a summary
Remote work Control from your phone with a quick command on Telegram

9. Conclusion

LobsterAI demonstrates that a full‑featured AI personal assistant can be built on open‑source tooling. From local file manipulation to remote scheduling, it covers the spectrum of everyday productivity tasks. With a robust architecture, sandboxed execution, and mobile‑friendly IM controls, it’s ready for both hobbyist experimentation and enterprise‑grade deployment.

Give it a try, fork the repo, and start building your own AI workflows today!