Easel – An Open-Source AI Social Media Content Workbench Jointly Developed by Zhejiang University and Peking University

1. What is Easel

Easel is an open-source AI social media content workbench jointly developed by Zhejiang University and Peking University, built upon the OpenClaw Agent framework. This platform connects the entire social media operation workflow through a single intelligent agent, covering the complete process from hot topic detection, topic planning, content creation, to multi-platform publishing and data analysis. By continuously remembering the account's positioning, style, and audience preferences through account profiling, combined with 112 executable Skills, the platform can continuously optimize its output quality as it is used, forming a truly "evolving" content collaboration system. The project uses the Apache 2.0 open-source license, with code hosted on GitHub, and has already received over 221 stars.

Image source: Official article
Image source: official article

Technical positioning and domain: Easel belongs to the domain of social media content automation driven by intelligent agents. Its core innovation lies in deeply integrating the generative capabilities of large language models with browser automation technology, forming an AI workbench capable of autonomously completing the entire social media operation workflow. Unlike traditional content management tools, Easel is positioned as an "AI Agent-native" content production system, rather than a simple editing assistant.

Development background: The project was jointly developed by the REAL Lab at Zhejiang University and the OpenDCAI Lab at Peking University. Both labs have deep technical expertise in natural language processing, multimodal learning, and intelligent agent systems. The development motivation stems from real-world challenges in current social media operations, such as fragmented content production pipelines, cumbersome cross-platform adaptation, and difficulties in maintaining consistent account styles. The goal is to integrate these scattered operational components into a coherent automated workflow using intelligent agent technology.

Core value: Easel solves the full-chain automation problem in social media operations, from inspiration to publishing. The platform ensures style consistency through account profiling, guarantees the timeliness of topics through a hot topic radar, directly generates publishable content in various formats using 112 executable Skills, and forms an optimization loop through data attribution. Its core value lies in upgrading AI from a "content advisor" to a "content executor," significantly reducing repetitive tasks and time costs in multi-platform operations.

Technical features: The platform is built on the OpenClaw Agent framework, enabling autonomous decision-making and tool calling. All 112 Skills are executable scripts rather than simple prompt templates, capable of directly generating text, images, audio, and video materials. In addition, the platform has implemented real publishing functionality through browser automation, supporting six major Chinese social media platforms, and has established an automatic feedback learning mechanism from publishing data to account profiling.

2. Key Features

• Account Profile System: Establishes account context through four core dimensions: positioning, style, audience, and boundaries. A single profile can be continuously used across multiple platforms. This system maintains a persistent memory of the account's publishing preferences and content boundaries, ensuring that AI-generated content always aligns with the account's tone and avoids style drift. Profile data continuously accumulates and updates with usage, forming a personalized model that becomes increasingly accurate over time.

• Hotspot Radar: Aggregates real-time trending topics from multiple platforms such as Weibo, Douyin, and Zhihu, and automatically filters relevant topics based on the account profile. This feature addresses the most time-consuming part of content creation—the topic selection process—by transforming a vast amount of hotspot information into personalized topic suggestions through intelligent filtering mechanisms, significantly enhancing content timeliness and relevance.

• Content Planning and Scheduling: Converts hot topics into a topic matrix, titles, scripts, and scheduling plans, all unified into a content calendar for management. This feature systematizes fragmented creative ideas, helping users establish an executable content matrix. It supports pre-planning of publishing schedules, ensuring the continuity and planning of account content updates.

• Multi-format Content Creation: Generates various publishable materials such as copy, posters, knowledge cards, voiceovers, AI videos, and short films. Based on 112 executable Skills, each content format has a dedicated professional generation process, ensuring output quality meets publishable standards, rather than just providing draft text.

• Multi-platform Intelligent Adaptation and Publishing: Automatically adapts master content to the format of each platform and publishes it directly to logged-in accounts after completing sensitive word and copyright checks. The system supports six major Chinese platforms: Xiaohongshu, Douyin, Kuaishou, Bilibili, WeChat Video Account. It automatically handles each platform's formatting requirements, tone, and aspect ratio, enabling one piece of content to be distributed across multiple channels.

• Data Attribution and Closed-loop Optimization: Automatically collects playback, interaction, and comment data, and feeds effective insights back into the account profile, forming a continuous learning loop for optimization. By analyzing data feedback from each platform, the system can identify which content strategies are effective and automatically adjusts profile weights, making subsequent content generation more aligned with the real preferences of the audience.

• Project-based Archiving and Management: Saves all outputs and metadata under the outputs/ directory by project, facilitating future modifications, retries, and reuse. This feature ensures that all creative assets are well-documented, preventing content from being lost in chat records. It also supports tracing and iterative optimization of historical projects.

3. How to Use

Easel's deployment and usage process is designed to be clear and straightforward. Here is the complete guide:

1. Environment Preparation: Ensure the operating system is Linux or macOS, and install Python 3.10+, Node.js 22.19+, and git. These version requirements are the foundation for running the OpenClaw framework and Easel. It is recommended to use the latest stable versions to avoid compatibility issues.

2. Clone the Repository: Execute git clone [email protected]:ZJU-REAL/Easel.git to pull the code to your local machine. It is recommended to clone the project into your working directory and ensure that your network environment can access GitHub normally.

3. Configure the API Key: Copy .env.example to .env and enter the LLM API Key (e.g., Anthropic). The API Key is the core credential for the Agent to perform tasks, and it should be kept secure. Do not commit it to a public repository.

4. One-Click Installation: Run bash setup.sh to automatically install Python dependencies, Node.js, and the OpenClaw framework. This script automates dependency management, reducing the likelihood of configuration errors.

5. Environment Check: Execute easel doctor and easel ping to confirm that the environment and gateway are functioning properly. These commands check whether all system components are intact and whether the API connection is smooth, ensuring that subsequent operations are not interrupted due to environmental issues.

6. Install Extensions: If you need audio and video capabilities and publishing functions, run pip install -e ".[media]" and playwright install chromium. The media extension requires additional runtime support, and Playwright's Chromium browser is a core component for automated publishing.

7. Create Account Profile: Copy the profiles/_template template or configure the account's positioning, style, and audience through the web interface. The quality of the profile directly affects the accuracy of all subsequent content generation. It is recommended to invest sufficient time in refining the profile description.

8. Launch the Workspace: Run easel web and open http://localhost:7860 in your browser to access the visual interface. The web interface provides a complete set of operation entry points, including hot topic browsing, content calendar, content creation dialogues, and the publishing center.

9. Discovery and Planning: Use the "Hot Topic Radar" to aggregate trending topics from various platforms and select topics. Use the "Content Calendar" to plan topics, titles, and scripts, forming an executable content matrix.

10. Creation and Publishing: Use dialogues or the easel skill command to let the Agent generate copy, cards, videos, and other materials. After completing sensitivity checks and copyright verification, as well as multi-platform format adaptation in the "Publishing Center," publish with one click.

4. Pros and Cons Analysis

5. Comparative Analysis with Similar Tools

To comprehensively evaluate Easel's technical positioning, the following comparison with Postiz and Buffer is provided:

Selection Recommendations: For individual bloggers, MCN agencies, and brand matrix operation teams targeting the Chinese market, Easel's autonomous content production capabilities and profile evolution mechanism can significantly enhance content production efficiency. Its deep integration with platforms such as Xiaohongshu, Douyin, and Bilibili provides a unique advantage. If the team has the technical capability to handle deployment and maintenance, Easel is a worthwhile open-source option to consider.

For cross-border e-commerce or overseas brand operation teams, Postiz's ability to publish on 25+ international platforms via API and its mature open-source community ecosystem are more appealing. However, the AGPL-3.0 license should be noted for its restrictions on commercial closed-source integration. The community's 35,000+ Stars indicates a wealth of documentation and third-party tutorial resources. If the enterprise requires fully managed commercial services and a highly reliable data reporting system, SaaS solutions like Buffer remain a safe choice, though they come with subscription costs and lack the autonomous content production capabilities driven by an Agent.

6. Editor's Summary

Easel demonstrates a clear innovative approach in its technical architecture by deeply integrating the OpenClaw Agent framework with an account profile system, achieving a complete automated closed-loop from content discovery to publishing attribution. Its design philosophy of 112 executable Skills is worth noting—compared to many current AI writing tools that only provide textual suggestions, Easel truly embeds AI capabilities into the content production execution layer, directly generating multi-format publishable materials such as images, text, audio, and video. This is relatively rare among similar open-source projects.

In terms of practical value, the platform has been deeply optimized for the Chinese social media ecosystem, supporting format adaptation and real publishing capabilities across major platforms such as Xiaohongshu, Douyin, and Bilibili. This provides direct productivity value for Chinese content creators. The profile-driven continuous learning mechanism is a standout differentiator, as it continuously optimizes generation quality through data feedback, evolving AI outputs from "generic templates" to "account-specific" content. This capability is still lacking in most current social media tools.

The project's open-source nature and academic background are notable advantages. The Apache 2.0 license removes licensing barriers for commercial integration, while the joint research and development background from Zhejiang University and Peking University provides credibility in terms of technical depth and sustainable iteration. The project is still in its early stages (221+ stars), with a relatively high deployment threshold. Its full coverage of Chinese platforms implies a dependency on automation components like Playwright—changes to platform pages or risk control strategies may affect publishing stability, which is a common development curve for open-source tools.

Easel's target audience is clearly defined: Chinese content creators with basic technical skills, bulk content production teams within MCN organizations, and brand entities requiring multi-platform matrix operations. For teams seeking a cost-effective, deeply customizable, and continuously evolving social media operations solution, Easel offers a compelling open-source option. Its future development is worth watching, particularly in terms of the richness of the Skill ecosystem, platform adaptation stability, and community building progress.

7. Application Scenarios

• Daily Operations for Individual Content Creators: After configuring their account profiles, creators can use Easel to automatically capture trending topics and generate tailored text, image, or video scripts for Xiaohongshu or Douyin. The platform automatically matches the content tone based on the profile's style and audience, allowing creators to simply confirm the final output in the publishing center. This significantly reduces the time required to produce content across multiple platforms on a daily basis.

• Multi-Platform Matrix Operations for Brands/Enterprises: When managing multiple platform accounts, Easel automatically generates platform-specific versions of content from a single master version, ensuring consistent tone and enabling bulk publishing. This reduces redundant work across platforms. The marketing team can focus on core creative ideas and brand strategies, while format adaptation and distribution execution are handled by the Agent.

• Bulk Content Production for MCN Agencies: Agencies can create individual profiles for each of their influencer accounts and use 112 Skills to generate diverse content formats in bulk. Content is archived by project for easy reuse and iteration. Each influencer account's unique style is preserved through their individual profile, enabling large-scale production while maintaining personalized characteristics for each account.

• Content Conversion for Academic/Knowledge-Based Accounts: Researchers can input their papers or specialized knowledge into Easel, and the Agent will automatically generate knowledge cards, explanatory articles, and video interpretations, transforming academic achievements into social media-friendly content. This scenario fully leverages the platform's long-text breakdown and multi-modal generation capabilities, helping research teams build academic influence.

8. FAQ

Q: What hardware requirements does Easel have?
A: Easel itself does not perform large model inference, as LLM calls are completed via a cloud API. Therefore, there is no mandatory requirement for a GPU. The core hardware requirements are sufficient RAM (recommended to be above 8GB) and disk space (approximately 10GB for dependencies and caching). It is recommended to use an SSD to improve the speed of Playwright browser and data processing.

Q: What are the most common errors during deployment?
A: The most frequently encountered issue is version mismatch in the environment, especially when Node.js versions are below 22.19 or Python versions are below 3.10, which can cause the setup.sh script to fail. The second common issue is the interruption of Playwright Chromium downloads due to network conditions; it is recommended to configure a domestic mirror source to resolve this. Running easel doctor can quickly identify problematic components.

Q: Is it necessary to manually update the account profile?
A: No frequent manual updates are required. The system automatically analyzes and aggregates data (such as playback, interaction, and comments) and updates the profile accordingly, enabling continuous evolution. It is recommended to periodically (e.g., weekly) check the update status of the "audience feedback" dimension in the profile, and manually adjust it only if there are significant changes in strategy.

Q: How does the platform avoid sensitive words or copyright risks in published content?
A: The publishing center includes a built-in quality gate mechanism, which consists of sensitive word detection (covering the review rules of major platforms), copyright verification (image and text source tracing checks), and SEO recommendations. Content that fails the gate will be intercepted and provided with modification suggestions. After passing, it will undergo a second check during the automated browser publishing process.

Q: Can multiple accounts be managed simultaneously while ensuring content remains distinct?
A: Yes. Each platform account is isolated through its independent account profile. The system automatically adjusts content style and format based on the target platform and profile information. By specifying the profile ID in the content creation conversation, the Agent will strictly follow the profile's positioning and red-line rules to generate content.

9. Project Links

• Project Website: https://zju-real.github.io/Easel/
• GitHub Repository: https://github.com/ZJU-REAL/Easel