First Hands‑On Review of JD.com’s Open‑Source JoyAgent‑JDGenie Multi‑Agent Platform

1. Installation

To quickly try the product, the author used a one‑click Docker approach. The steps are:

<code>1. git clone https://github.com/jd-opensource/joyagent-jdgenie.git
2. Manually update <code>genie-backend/src/main/resources/application.yml</code> with <code>base_url</code>, <code>apikey</code>, <code>model</code>, <code>max_tokens</code>, <code>model_name</code> etc. When using DeepSeek set <code>max_tokens: 8192</code>.
   Update <code>genie-tool/.env_template</code> with <code>OPENAI_API_KEY</code>, <code>OPENAI_BASE_URL</code>, <code>DEFAULT_MODEL</code>, <code>SERPER_SEARCH_API_KEY</code>. For DeepSeek set <code>DEEPSEEK_API_KEY</code>, <code>DEEPSEEK_API_BASE</code> and change <code>DEFAULT_MODEL</code> to <code>deepseek/deepseek-chat</code> (replace all <code>${DEFAULT_MODEL}</code> accordingly).
3. Build the Docker image:
   <code>docker build -t genie:latest .</code>
4. Run the container:
   <code>docker run -d -p 3000:3000 -p 8080:8080 -p 1601:1601 --name genie-app genie:latest</code>
5. Open a browser at <code>localhost:3000</code> to access the UI.
<p>The author’s environment was Ubuntu 22.04 LTS (jammy). Two notable issues were encountered:</p>
<ul>
<li>When editing <code>application.yml</code>, the <code>model_name</code> field must be replaced for the chosen model (e.g., <code>deepseek-chat</code>).</li>
<li>During Docker image building, the Debian package source in the Dockerfile used the <code>bookworm</code> repository, but the base image <code>python:3.11-slim</code> runs Debian 13 (trixie). The Dockerfile was manually edited to point to the correct <code>trixie</code> sources.</li>
</ul>
<p>After fixing these points, the image built successfully and the UI displayed a clean homepage.</p>
<h2>2. Case Test – Code Architecture Analysis</h2>
<p>The first demo used the built‑in "code architecture analysis" case. The prompt was:</p>
<blockquote>"Please analyze the open‑source joyagent‑jdgenie repository on GitHub and provide an analysis."
</blockquote>
<p>When executed, the left pane showed the step‑by‑step reasoning while the right pane displayed real‑time results. The author noted that JDGenie performed extensive searches on GitHub and the broader web, which was visualised in the UI.</p>
<p>After about eight minutes, JDGenie produced an analysis report available in both HTML and Markdown formats. The report correctly outlined the project but contained an error stating the project was open‑sourced in March 2021, which the author identified as inaccurate.</p>
<h2>3. Case Test – Physics Animation Generation</h2>
<p>The second demo reproduced a classic Manus task: creating a series of clear, accurate animation demos for a high‑school physics lesson on momentum conservation, delivered as a simple HTML presentation.</p>
<p>Manus generated a satisfactory set of collision animations with adjustable parameters. JDGenie was then asked to perform the same task. It produced an HTML page with a reasonable layout and styling, but several elements failed to render and the animation itself did not work, as shown in the screenshots.</p>
<p>The author concluded that while JDGenie’s UI and report generation are promising, its output quality still lags behind Manus, especially for dynamic content.</p>
<h2>4. Summary</h2>
<p>Overall, JDGenie offers a higher level of completeness compared with other open‑source multi‑agent projects, covering both front‑end UI and back‑end services. However, the generated results still have noticeable gaps, and the project, being newly open‑sourced, requires further community iteration to reach parity with more mature solutions.</p>