A desktop AI beaver dam for the Xiaohongshu content mill

What it does Beav—formerly RedBox—is a macOS and Windows desktop app that turns a local machine into a content creation control tower for Xiaohongshu influencers and e-commerce teams. A browser extension funnels posts, videos, and articles from Xiaohongshu, YouTube, and WeChat into a built-in knowledge base, where AI agents draft manuscripts, generate images with GPT-image-2, edit video, and assemble covers. It is essentially a domain-specific operating system for social media production, bundling asset management, automation, and team collaboration into one native application.

The interesting bit The project treats content creation like a CI/CD pipeline: a “wander” mode randomly shuffles your knowledge base to spark ideas, a “RedClaw” runner executes scheduled and long-period tasks in the background, and a subject library pins reusable characters, products, and scenes across drafts. It also generates multi-language product detail pages for dozens of cross-border platforms—from Tokopedia to Amazon EU—suggesting its ambitions extend far beyond a single app.

Key highlights

• Browser plugin captures Xiaohongshu posts, YouTube videos, WeChat articles, and generic web pages directly into a structured local knowledge base.
• Integrated media suite using GPT-image-2 for batch image sets, plus text-to-video generation and AI video clipping with automatic subtitle recognition.
• RedClaw agent system handles continuous, scheduled, and long-running automation tasks without manual babysitting.
• Supports multi-language product detail page generation for cross-border e-commerce platforms including Shopee, Lazada, Temu, and Amazon EU.
• Team collaboration features with member profiles, shared knowledge, and group chat-style workflows.

Caveats

• The app requires external AI API keys (Endpoint/Key/Model) and is not fully offline; the “localization” branding refers to the desktop-native architecture, not local LLM inference.
• The primary README and interface are overwhelmingly Chinese; English documentation exists but appears secondary, so non-Chinese speakers may hit friction.
• The MIT-NC license explicitly forbids commercial use, which sits awkwardly alongside features built for monetized content and cross-border e-commerce teams.

Verdict Chinese-speaking content creators and small e-commerce teams managing Xiaohongshu and short-video pipelines should take a look; developers seeking a general-purpose AI agent framework or a fully open commercial tool will find the narrow domain focus and non-commercial license limiting.