LobsterAI: NetEase Youdao open-sources a desktop AI agent that works in Office

LobsterAI is an open-source desktop AI agent built by NetEase Youdao, designed to actually carry out hands-on work on your local computer. Early community feedback has described it as an out-of-the-box alternative to tools like OpenClaw: you just double-click the installer, skip the tedious terminal setup, and get right to work without losing execution quality.

What it does

While most AI chatbots stop at generating text, LobsterAI is built to finish the actual deliverable. You simply point it at a local project folder, and it can analyze spreadsheets, write reports, execute scripts, and navigate the web.

Document workflows are native. It opens Word, Excel, and PowerPoint files directly inside the application. You can watch the agent update data in real time, or step in and make manual tweaks side by side. It also includes an artifact preview panel for HTML, Mermaid diagrams, Markdown, images, and video.

Remote delegation is another highlight. LobsterAI connects directly with everyday chat platforms, including WeChat, Lark, DingTalk, Telegram, and Discord. Once configured, you can leave your desktop running, send a prompt from your phone while away from your desk, and let the agent handle the heavy lifting.

For routine tasks, the software supports scheduled jobs. You can schedule it to run morning research briefs, check web analytics dashboards for anomalies, or compile daily wrap-up reports automatically.

How it stays safe

Running an autonomous agent on your personal machine raises valid security concerns. LobsterAI addresses this with a sandbox approach similar to Claude Cowork. Before taking any sensitive action, such as altering a file, executing a shell command, or making network requests, the interface halts and requires explicit user approval.

Under the hood, it features durable memory and scheduled "Dreaming" runs that consolidate interaction patterns across sessions. It also supports local and remote MCP connectors for custom toolkits.

Model flexibility is broad. After signing in, you can tap into built-in Chinese models like DeepSeek, GLM, Kimi, and Qwen, or bring your own API keys for OpenAI, Anthropic, Gemini, or local setups via Ollama and LM Studio.

LobsterAI is fully open-sourced under the MIT license on GitHub. Pre-built binaries are available for Windows x64 and macOS 13 or later, supporting both Apple Silicon and Intel chips.

Why it matters

We think LobsterAI makes a very smart move by focusing on distribution and workflow friction rather than competing on core foundational models. Bringing native Office editing, sandboxed approvals, and mobile chat triggers into an easy-to-install desktop app significantly lowers the barrier to everyday local automation.

If you deal with piles of spreadsheets or want to assign tasks to your computer while on the go, it is definitely worth testing right away. Keep in mind that full computer-use tasks still depend heavily on the underlying model's reasoning limits, so you will still want to keep an eye on safety approvals during complex workflows.