lobsterai

LobsterAI is an open-source desktop AI agent from NetEase Youdao, distributed under the MIT License and packaged as an Electron application for macOS and Windows. It is positioned as an all-scenario office assistant: rather than only chatting, it operates inside the user's real working environment, touching local files, terminal commands, browser workflows, documents, spreadsheets, slides, IM channels, scheduled jobs and project workspaces. The project separates two layers. Cowork is the product and session layer inside the desktop app, holding local persistence, permissions, UI state, artifacts, agents, memory and IM bindings. OpenClaw is the runtime and gateway underneath, responsible for agent execution. Integration code such as openclawEngineManager, openclawConfigSync, openclawRuntimeAdapter and coworkEngineRouter translates LobsterAI state into OpenClaw runtime behavior. Key capabilities described in the README: - Desktop Cowork sessions: long-running agent tasks against local projects and files, with streamed progress, session history, rendered tool output, and approval prompts before sensitive actions such as file operations, terminal commands or network access. - Multi-agent workflows: user-created agents with their own identity, model choice, skills, working directory, enabled state and IM bindings, alongside a general Main Agent. - Expert Kits: installable scenario-oriented bundles that package capability selections and references, usable together with individual skills. - Skills: 28 built-in skills configured in SKILLs/skills.config.json, covering web search, Word documents, spreadsheets, PowerPoint, PDF processing, Remotion video generation, browser automation, image and video generation, stock research, content writing, email, weather and skill creation. - MCP servers: external tools and data sources connected through Model Context Protocol, stored locally and synced into OpenClaw when enabled. - Scheduled tasks: recurring work created by conversation or through a UI, for example daily news digests, inbox summaries, website monitoring and weekly reports. - IM remote control: reaching the desktop agent from WeChat, WeCom, DingTalk, Feishu/Lark, QQ, Telegram, Discord, NetEase IM, NetEase Bee, POPO and email, with multi-instance platforms able to bind different accounts or channels to different agents. - Rich artifacts: preview and management of generated HTML, SVG, images, video, Mermaid diagrams, code, Markdown, text, documents and local service artifacts inside the app. - Local memory and data: sessions and app data in local SQLite; OpenClaw workspace memory uses files such as MEMORY.md, USER.md, SOUL.md and daily notes so preferences and project context persist across sessions. Example prompts in the README include building a local inventory system from Excel data, analyzing a spreadsheet into a visual dashboard, researching a market and producing a presentation, automating morning browser checks of an ads dashboard, screening resumes against a job description, and a weekday 9 AM AI news digest. Architecture: the renderer uses React, Redux Toolkit and Tailwind with artifact renderers and configuration UIs; the Electron main process handles lifecycle, IPC, SQLite persistence, auth, logging, OpenClaw startup, runtime repair, skill sync, IM gateways and artifact services. Building from source requires Node.js >=24.15.0 <25, npm >=11.17.0 <12, git and pnpm; the first run builds a pinned OpenClaw runtime from a sibling checkout. Development commands include npm run electron:dev:openclaw, npm run electron:dev, npm run build, npm run compile:electron, npm test and npm run lint. Packaging scripts (dist:mac, dist:win, dist:linux and variants) build installers per host OS, with Windows-specific steps for portable Python and Git. A separate DeepSeek Harness (dsh) runtime has its own build, pack, verify and end-to-end scripts, with per-platform archives verified by digest. Security notes: renderer windows use context isolation, disabled Node integration and sandboxing; renderer-to-main access goes through preload IPC APIs; sensitive tool actions are permission-gated and logged; app data lives in lobsterai.sqlite under Electron userData, with OpenClaw state, workspace memory, generated config and gateway logs under userData/openclaw. Community support is offered through a WeChat group, GitHub issue templates and pull request guidance. The project is built and maintained by NetEase Youdao.