Easel: An AI Agent That Runs Your Whole Content Operation, Not Just Drafts It

Easel's GitHub star count went from 417 to 731 in two days, a jump big enough that it is worth asking what the project actually does before writing it off as another repo having a viral moment. The answer: Easel, an open-source project from ZJU-REAL, is a standing AI content agent that discovers what is trending, writes and produces the content, publishes it natively to the platform, and comes back later to check what actually worked, all without a chat window to re-brief every morning. That is a genuine re-acceleration too, not just a first spike, after two prior scan windows saw it grow without a comparable jump.

What Easel actually does

Easel's own framing is "from one idea to completion: discover, plan, create, publish and review," and the pipeline maps directly onto that. It watches for trending topics on the platforms it targets, plans a content calendar around what it finds, produces the actual content (text, images, video, and audio, depending on the format), logs into the target platform through a real browser session via Playwright, and publishes natively rather than through a workaround. After publishing, it tracks performance and feeds what worked back into the next round of trend discovery, so the loop compounds instead of resetting every time.

The platforms it targets are Chinese-first: Xiaohongshu, Douyin, Kuaishou, Zhihu, Bilibili, and WeChat Video. There is no English-language platform support (no X, Instagram, or YouTube) as of this writing, which matters more here than it would for a general-purpose tool, since the entire value proposition is native publishing on the platforms it knows.

Setting it up and running it

Easel installs through a guided setup script (bash setup.sh on Linux and macOS, a PowerShell equivalent on Windows) that checks for the dependencies it needs: Python 3.10 or later, Node.js 22.19 or later, Git, and FFmpeg, then builds a Python virtual environment and installs both the Python and Node package sets. Under the hood it runs on the OpenClaw agent framework with a FastAPI backend and a React frontend, and it needs an LLM to actually think, Claude, GPT-4, or a compatible model through the same interface. Once set up, easel web opens a local interface on localhost:7860, and from there the agent runs its own loop rather than waiting for a new prompt each time. It is Apache 2.0 licensed, genuinely permissive, and free.

The account-profile piece is worth calling out specifically: Easel maintains a persistent profile per account, so its output stays consistent in voice and positioning across sessions instead of drifting the way a fresh chat conversation would each time it is re-briefed.

The part its own documentation is honest about

Easel's own README carries a direct warning about Xiaohongshu specifically: the platform may detect automated posting, and the risk runs from a verification challenge to rate-limiting to an outright account restriction. Its own guidance is to preview before publishing there rather than let it run unattended, and it says the other five platforms operate normally by comparison. That is a meaningfully different posture from a tool that stays silent on the account risk of automated posting, and it means the "publishes it across platforms end to end" pitch has a real asterisk on the platform where the audience is arguably largest.

The broader caveat sits above that specific warning: this is a young project, created in late August 2026 and past 900 stars and 125 forks already, not a mature, battle-tested system with years of edge cases already found by other people running it. An agent that logs into a real account and posts on your behalf, on any platform, carries the same category of risk this site has flagged before with browser-automation agents: the account's standing rests on the platform tolerating a level of automation its own terms of service were not written to expect.

Who Easel fits, and who it doesn't

Easel is a real fit for a solo operator or a very small team running content specifically on Xiaohongshu, Douyin, Zhihu, Bilibili, Kuaishou, or WeChat Video, who wants an AI content agent that runs the loop on its own rather than one they re-brief in a chat window every day. It is a weaker fit for anyone whose audience lives on X, Instagram, LinkedIn, or YouTube, since none of those are supported targets today, and for anyone unwilling to accept real account risk in exchange for native, unattended publishing. For that second group, the discovery-and-drafting half of the loop is still worth studying even without adopting the publishing half: an agent that closes the loop from trend to performance data, rather than stopping at a draft, is a genuinely different shape from most content tools on the market, whether or not this specific implementation is the one to run in production yet.

The honest read: Easel is not a polished, enterprise-ready content department in a box. It is a fast-moving, technically real open-source agent that automates the full content loop on a specific set of Chinese platforms, with its own developers flagging exactly where the automation is riskiest. For the audience it targets, that combination of speed and honesty about the risk is worth a real look before dismissing it as another trend-chasing repo.

Sources: ZJU-REAL/Easel on GitHub, accessed September 10, 2026.