Easel Review 2026 — An Open-Source AI Content Workspace That Discovers, Creates and Publishes Social Media Posts Across Six Chinese Platforms
✅ Pros
- • A full-loop social media content agent: five connected workflows (Discover, Plan, Produce, Publish, Attribute) in one continuous flow, so trend discovery, topic scoring, content calendar, creation, publishing and performance attribution aren't bolted-together point tools
- • Profile-driven, not one-shot generation: every account gets its own profile (positioning, style, audience, platforms, preferences and red lines, long-term memory) under profiles/<name>/ that persists across platforms, sessions and content cycles
- • 112 executable skills with real scripts, not feature-list promises: image/card/poster/infographic generation, TTS, multi-role dubbing, voice cloning, AI music, subtitles, video editing, short-drama, and publishing skills write real deliverables to outputs/, with a skill-function-mapping doc
- • One source, many platform forms: a single topic can be adapted into Xiaohongshu cards, a short video, a Zhihu long-form article or a short post, each respecting the platform's format and length conventions
- • Real publishing and attribution loop: login/adaptation/publishing workflows for Xiaohongshu, Douyin, Kuaishou, Zhihu, Bilibili and WeChat Channels, with quality gates, risk checks, publishing checklists, and performance data fed back into the account profile
- • Research-backed and open: from Zhejiang University's REAL Lab and Peking University's OpenDCAI Lab, Apache-2.0, with a Web workspace (FastAPI + React) plus CLI, and an isolated OpenClaw profile so it doesn't clobber an existing setup
⚠️ Cons
- • Chinese-platform-first: publishing targets Xiaohongshu, Douyin, Kuaishou, Zhihu, Bilibili and WeChat Channels — creators focused on Western platforms (Instagram/TikTok/X) won't get the publishing loop, and the docs are primarily Chinese (English README and project page exist)
- • Xiaohongshu automation risk is real and disclosed: the README warns that automated publishing may trigger verification, reach limits or account risk controls, and recommends previews and human-confirmed publishing — so the flagship platform is exactly where you must be most careful
- • Fresh and fast-moving: created 2026-08-28, first release v0.1.0 on 2026-08-31, ~200 stars/20 forks at review time — impressive for a week, but no community track record and APIs may shift
- • Requires an LLM key and optional media providers: minimal config needs ANTHROPIC_API_KEY (Claude model), and video/music/cloud-voice capabilities depend on external provider keys and services — it's an orchestrator, not a self-contained media suite
- • Heavier setup than a SaaS alternative: Linux/macOS, Python 3.10+, Node.js 22.19+, OpenClaw, plus optional `.[media]` extras, Playwright Chromium and FFmpeg for image/audio/video and browser publishing
- • Browser publishing depends on logged-in platform accounts and platform UI stability — the README itself notes real publishing can be affected by verification, permissions, risk controls and UI changes
Chinese social-media creators, brand content teams and studios (especially Xiaohongshu/Douyin/Zhihu operators) who want an open, self-hosted, profile-driven agent that handles the whole content loop — trend discovery to planning to card/video production to (carefully) publishing — and researchers exploring agentic social content workflows, given it's Apache-2.0 and built on OpenClaw
Free, Apache-2.0, open source (Python 3.10+). Requires an Anthropic-compatible LLM key (minimal .env: ANTHROPIC_API_KEY + CLAUDE_MODEL, default anthropic/claude-sonnet-4-6); optional paid providers for AI video, music and cloud voice; platform publishing needs your own logged-in accounts
The Pitch: A Content Partner That Remembers
On August 28, 2026, Zhejiang University’s REAL Lab (with Peking University’s OpenDCAI Lab) released Easel — an open-source “content workspace for social media creators” that takes a different stance from most AI writing tools. Instead of answering “how should I do this?”, Easel’s agent actually makes the content: it connects an OpenClaw agent, account profiles, content skills, and real media tools, then produces and archives finished posts — with direct or on-demand publishing when you want it.
The framing is “a content partner that remembers.” Each account has an independent profile — positioning, style, audience, platforms, preferences and boundaries (red lines), and long-term memory — that persists across platforms and sessions. Outputs get more on-brand over time because performance data flows back into the profile. In its first week the repo passed 200 stars and 20 forks, with a v0.1.0 release on August 31.
The Five-Layer Workflow
Easel’s architecture is five connected workflows, and the pipeline is the product:
- Discover — aggregate hot lists and industry news across Weibo, Douyin, Zhihu, Bilibili, Baidu and Toutiao; vertical trend research, content-gap analysis, event calendars, platform algorithm updates, competitor research, RSS aggregation and UGC discovery.
- Plan — turn opportunities into topics: positioning analysis, audience profiles, persona and voice, account diagnosis, topic matrices with scoring, hooks and titles, scripts, storyboards, a content calendar, and series/campaign/livestream planning.
- Produce — generate the actual assets: social copy and Xiaohongshu notes, long-form articles and novels, paper explainers, quote/knowledge cards, posters, infographics, charts, mind maps, memes and AI images — plus audio and video: TTS, multi-role dubbing, voice cloning, AI music, denoising/mixing, transcription, subtitles and translation, video editing, clipping, highlights, format conversion and green screen.
- Publish — adapt titles, copy, aspect ratio and media per platform, run quality gates and risk checks (sensitivity, copyright), then publish through logged-in accounts for six platforms: Xiaohongshu, Douyin, Kuaishou, Zhihu, Bilibili and WeChat Channels.
- Attribute — collect views, engagement, comments and content performance, run postmortems and ROI analysis, and preserve what works back into the account profile.
The 112 skills are mapped in a capability document (docs/skill-function-mapping.md), and the README’s showcase shows real outputs: paper-explainer knowledge cards, novels across genres (cultivation comedy, horror, romance, suspense), lifestyle and meme cards, and finished videos including a talk-show drama and an AI character mukbang clip.
Profiles, Workspace and the OpenClaw Foundation
Easel is built on OpenClaw — it manages an isolated easel OpenClaw profile with its own workspace, so installing it doesn’t overwrite an existing OpenClaw configuration. The Web workspace (FastAPI backend + React frontend, default http://localhost:7860) manages sessions, assets, accounts, profiles, a content library and publishing states; the gateway defaults to port 18789. A CLI mirrors the essentials: easel web, easel chat (multi-turn terminal chat with profile selection), easel skill <name> -i "..." [-p profile] to run a single skill directly, plus easel doctor and easel ping for environment checks.
Profiles live under profiles/<name>/ with six dimensions and can be created from a _template or in the Web workspace. Content projects are archived under outputs/ with source material, intermediate files, metadata and deliverables together — so revision, retry and republishing don’t scatter across chat history. Setup is bash setup.sh plus a minimal .env (an Anthropic-compatible key); optional pip install -e ".[media]", Playwright Chromium and FFmpeg unlock image/audio/video and browser publishing.
Honest Limitations
The biggest caveat is scope: Easel is built for the Chinese social ecosystem. Its publishing loop targets Xiaohongshu, Douyin, Kuaishou, Zhihu, Bilibili and WeChat Channels — creators working Instagram/TikTok/X get the discovery/creation half but not the publishing loop, and most documentation is Chinese (an English README and project page exist, and the UI/skills work regardless of language). The second caveat is disclosed by the authors themselves: automated Xiaohongshu publishing risks verification, reach limits or account risk controls, so they recommend preview and preflight checks with human-confirmed publishing — meaning the flagship channel is where automation must be most conservative. Third, it’s a week-old v0.1.0: no community track record, and media capabilities depend on external providers you configure (video, music, cloud voice). Finally, it’s an orchestrator, not a turnkey media suite — local image/audio work needs the media extras and FFmpeg, and browser publishing depends on logged-in accounts and platform UI stability.
Verdict and Who It’s For
Easel is the most complete open-source answer we’ve seen to the “content operations loop” problem for Chinese platforms: one profile-driven agent that discovers trends, plans a calendar, produces cards/videos/long-forms with genuinely executable skills, publishes with quality gates, and learns from performance. For Xiaohongshu/Douyin/Zhihu creators and brand teams who want a self-hosted, Apache-2.0 alternative to SaaS content tools — and for researchers interested in agentic social content — it’s a Silver-tier pick with real showcase outputs, provided you respect the Xiaohongshu automation caution and bring your own LLM/media keys.
Review based on public repository contents, README (EN), project page, and repository metadata as of 2026-09-03. Star/fork counts reflect a project six days old at review time.