Easel: an OpenClaw agent that turns one topic into posts for six Chinese platforms
An open-source AI agent for social media — discover trends, create content, publish everywhere, and learn what works across Xiaohongshu, Douyin, Zhihu, Bilibili, and more.🎨一个开源的 AI 社交媒体智能体——发现热点趋势、创作内容、一键发布至各大平台，并学习分析哪些内容真正有效，覆盖小红书、抖音、知乎、哔哩哔哩等平台。
At a glance
- What is it?
- Easel is an Apache-2.0 Python agent from ZJU-REAL that chains trend discovery, content creation, publishing and attribution across Xiaohongshu, Douyin, Kuaishou, Zhihu, Bilibili and WeChat Video Accounts. The repository ships 112 skills with runnable scripts, but the README itself flags automated posting to Xiaohongshu as risky.
- Who is it for?
- Adopt Easel if you already operate accounts on several Chinese platforms, want generated assets written to disk under outputs/ rather than pasted into a chat window, and are willing to run the Web frontend instead of the CLI. Do not adopt it if your only channel is Xiaohongshu and you expect unattended publishing, because the README warns the platform may detect automation and trigger verification, rate limits or account controls.
- Can I use it commercially?
- Yes. Apache-2.0 is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 1 day ago.
- What is it written in?
- Mainly Python, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
The gap Easel targets: one topic, six platform formats, no shared memory
A creator running accounts on Xiaohongshu, Douyin, Zhihu and Bilibili does the same job four times. The topic is fixed, but the card dimensions, the title length, the tone and the posting checklist all change per platform. Most AI writing tools stop at text and forget the account. Easel's pitch is that the account is the input: the README describes a profile carrying positioning, style, audience, platform, preferences and memory, and says one profile can be reused across platforms and sessions. That is the specific problem. It is aimed at individual creators and small operations teams who already publish in Chinese on more than one network and who want the intermediate files, not just the caption. The repository is maintained under ZJU-REAL and OpenDCAI Lab, licensed Apache-2.0, and the only release listed is v0.1.0 from 2026-08-31.
Five stages, one agent: how discovery, planning, creation, publishing and attribution connect
The README lays out a five-layer workflow: discover, plan, create, publish, attribute. Discovery aggregates hot lists from Weibo, Douyin, Zhihu, Bilibili, Baidu and Toutiao, plus industry news, competitor activity and user discussion, then filters for fit with the account. Planning converts an opportunity into a topic, title, script and calendar entry. Creation produces copy, cards, posters, infographics, audio and video, and the README states these skills ship with runnable scripts whose output is written into outputs/. Publishing adapts title, body, aspect ratio and media to each platform, runs a pre-publish check, and sends to a logged-in account. Attribution reads plays, interactions and comments, then writes what worked back into the profile. The loop is the design point: the profile is both the input to generation and the destination for results. Whether that feedback loop produces measurably better content is not something the README demonstrates, and I cannot confirm it without running the pipeline.
OpenClaw is the runtime, and the 112 skills are the surface area
Easel is not a standalone model wrapper. The README labels it OpenClaw-powered, and describes the architecture as connecting the OpenClaw Agent, account profiles, content skills and real media tools. The skill count is given as 112, with a mapping document at docs/skill-function-mapping.md. Those skills are grouped by stage: discovery skills for trend research and content-gap analysis, planning skills for positioning, audience profiles, topic matrices and calendars, text and visual skills for notes, long articles, quote cards, knowledge cards, posters, infographics and AI image generation, audio and video skills for text-to-speech, multi-role dubbing, voice cloning, subtitles, editing, long-video slicing and aspect-ratio conversion, and finally publishing and attribution skills covering quality gates, sensitive-content and copyright checks, platform format adaptation and account data retrieval. The README is explicit that these are executable scripts rather than a feature list. That claim is the one worth testing first, because a skill that shells out to a media tool you have not installed will fail at the point of use, not at install time.
Getting it running: Python 3.10+, the Web frontend, and where artifacts land
The README states Python 3.10+ and recommends the Web frontend over the CLI, saying the frontend provides the full set of session, asset, account, profile, content library and publishing management capabilities. It does not print a pip install line, a docker command or a config file in the excerpt available here, so I cannot give you an exact bootstrap sequence without inventing one. What the material does establish: skills execute scripts and write finished artifacts into outputs/, content is archived per project with assets, intermediate files and metadata, and publishing targets accounts you have already logged into on Xiaohongshu, Douyin, Kuaishou, Zhihu, Bilibili and WeChat Video Accounts. The practical implication is that setup is not just a Python install. You are wiring an agent runtime to media tooling and to live platform sessions. Treat the first run as an integration exercise: pick one skill, confirm its script executes and its output appears under outputs/, then add the publishing path.
The Xiaohongshu warning is the most important line in the README
The README's own usage note says to be cautious about automatic publishing to Xiaohongshu, because the platform may detect automated operation and this carries risk of verification, rate limiting or account restrictions. It recommends using preview and pre-publish checks and having the user confirm before publishing manually. It adds that other platforms behave normally, though it does not explain what normal means or what detection mechanisms exist elsewhere. This is a real limitation, stated by the maintainers rather than inferred. It also tells you something about the design: the pre-publish quality gate, sensitive-content check and copyright check exist partly because the publish step is the irreversible one. If your workflow depends on unattended posting to Xiaohongshu, Easel is the wrong tool by the project's own account. The same caution applies to any platform that tightens automation rules after v0.1.0; nothing in the repository guarantees the six supported integrations stay working.
Where Easel stops and a general agent framework begins
The obvious comparison is a general agent framework such as LangChain or a plain OpenClaw agent with hand-written tools. The difference is not the model call. It is that Easel pre-packages the domain: the account profile schema, the platform format rules, the calendar, the publish checklist and the attribution write-back are all built in, and the skills are numbered and mapped in docs/skill-function-mapping.md. With a general framework you would assemble the same pipeline yourself and own every platform adapter. The trade is flexibility for opinion. Easel decides that content is organized by project, that profiles persist across sessions, and that publishing is a gated step. If your process does not match that shape, for example if you publish through an agency tool or need a platform Easel does not support, the built-in adapters become something you work around rather than with. There is no plugin interface described in the available material, so extending to a new platform likely means writing a skill in the same style as the existing ones.
Licence, release cadence and what maintenance actually costs you
Easel is Apache-2.0, which permits commercial use and modification and includes an explicit patent grant, with the usual requirements to retain notices and state changes. That is a permissive choice and it matters if you plan to run this inside a paid content operation. I am not a lawyer and this is not legal advice; read LICENSE before shipping. On maintenance: the only release in the material is v0.1.0, published 2026-08-31, with the last push to main on 2026-09-10. A single early release means the API surface, skill scripts and profile schema can still move. Your upgrade cost is concentrated in two places. First, the skill scripts, because media tooling versions change and a script that worked against one encoder may not work against the next. Second, the platform adapters, which track login flows and format rules on six networks you do not control. Budget for re-testing the publish path after each upgrade, not just after each Easel release.
Who should install it, and the first thing to check
Install Easel if you run multiple Chinese social accounts, you want generated cards, audio and video saved as files under outputs/ with project metadata, and you are comfortable treating publishing as a reviewed step rather than an automatic one. Skip it if you publish only to Xiaohongshu, if you need a platform outside the six listed, or if you want a stable API to build against today; v0.1.0 does not offer that. The first thing to verify after cloning is narrower than a full smoke test: open docs/skill-function-mapping.md, pick the one skill your workflow cannot do without, run its script, and confirm the artifact lands in outputs/. If that works and the publish step stays manual, the rest of the pipeline is worth the setup time.
Editorial conclusion
Adopt Easel if you already operate accounts on several Chinese platforms, want generated assets written to disk under outputs/ rather than pasted into a chat window, and are willing to run the Web frontend instead of the CLI. Do not adopt it if your only channel is Xiaohongshu and you expect unattended publishing, because the README warns the platform may detect automation and trigger verification, rate limits or account controls. Before committing, verify two things yourself: that your Python environment is 3.10 or newer and that every skill script your workflow depends on actually runs against the media tools you have installed, since the README lists capabilities but not a per-script dependency manifest.
Community notes