ShoppingAgent Order Analyst
Captures order pages, extracts their contents with a configured model, and exports the results to Excel.
The README discloses browser automation, screenshots, model analysis, Excel export, and API-key configuration, while button-driven actions provide limited evidence of user confirmation. It does not document least-privilege controls, domain restrictions, screenshot or order-data destinations, key protection, dependency auditing, rollback, or export recovery. Repository naming and an author contact offer thin attribution, but the license and verified publisher identity are unknown.
Dependency categories, startup commands, and a troubleshooting FAQ provide basic deployment guidance, but dependencies are not locked and structured failure behavior or compatibility validation is absent. The prerequisite requires Chrome 140+, while the FAQ says 112+, and the Shopping Agent documentation is combined with legacy chatbot environments, reducing self-consistency.
The audience and workflow for capturing order backends, batch analysis, and Excel export are reasonably clear; server, model, and browser settings are configurable, and URL entry plus explicit capture and analysis buttons make triggering fairly precise. Deductions reflect missing boundaries for supported sites, authentication, anti-automation behavior, model capability, and scale, together with contradictory Chrome requirements and legacy Ubuntu 18.04/Python 3.6 constraints.
The README has useful navigation and sections for overview, setup, configuration, structure, workflow, and FAQ. Installation is adequate for an initial attempt, but lacks an environment template, verified driver location, and broad platform guidance. ShoppingAgent, Shopping Agent, and chatbot naming is mixed, while execute/excute spelling is unstable. The roadmap supplies limited version history but is dated and includes a tentative “Maybe”; no license is supplied, and limitations and maintenance ownership are only indirectly represented by the FAQ, roadmap, and QQ contact.
Batch screenshot analysis, order-data aggregation, and Excel export produce a plausibly usable artifact and offer meaningful value over manual transcription. Scores are reduced because no output schema, sample workbook, accuracy evidence, throughput, review workflow, model/API cost, or automation-maintenance cost is supplied, leaving cost-benefit claims weakly supported.
Most claims can be located in the README's feature, workflow, and dependency sections, and roadmap language distinguishes some planned work from current features. However, no code, tests, sample exports, configuration files, or independent source corroborates the Shopping Agent claims; the displayed results concern only the legacy Seq2Seq chatbot. Capability and compatibility statements therefore remain largely asserted.
- Browser automation accesses user-supplied order backends and stores screenshots. Before using real orders, establish allowed domains, credential isolation, screenshot location, retention, and deletion procedures.
- API-key storage and log redaction are undocumented. Do not use production credentials until file permissions, frontend exposure, and logging behavior are verified.
- The README gives conflicting Chrome minimums of 140 and 112; deployment should follow verified runtime and driver compatibility rather than the document alone.
- Dependencies are described only with minimum versions, without a lockfile, security audit, or license information. Independently audit and pin them before deployment.
- Excel output has no field specification, accuracy evidence, or human-review policy and should not directly drive financial, fulfillment, or other high-impact decisions.
What does this agent do, and when should you use it?
This repository combines ShoppingAgent, an order-processing application, with an older trainable Chinese Chatbot project. ShoppingAgent runs a FastAPI backend and exposes a Streamlit interface for chat commands, screenshots, result review, and configuration. Selenium opens order pages and captures full-page, element-level, or manually selected regions, which a configured language model analyzes in batches. Extracted order data appears in the interface and can be written to local Excel files. Deployment is self-hosted and requires Python, Chrome, a compatible ChromeDriver, model settings, and an API key, although the documentation does not identify a supported model provider.
A user submits an order-page URL through the Streamlit interface, and browser automation opens the page. The application can take a full-screen capture, target an element, or obtain a full-page image from which the user manually selects a region. It batch-analyzes collected screenshots with the configured model, presents the extracted order information in a side panel, and exports the results through pandas and openpyxl to Excel. The documented backend command is python -m app.main; the frontend starts with streamlit run app/frontend.py and is available at http://localhost:8501. The repository also contains a Chinese Seq2Seq Chatbot workflow in which data_utls.py preprocesses a corpus, execute.py runs training, and app.py supplies a visual chat module, with an additional Horovod command for distributed training.
- E-commerce operators whose administration portal cannot export orders directly can capture the pages and consolidate extracted fields into Excel.
- Teams processing many order screens can collect multiple screenshots and submit them for batch analysis instead of copying each order manually.
- Analysts working with portals that lack a documented data API can use browser automation and image-based extraction as an alternative input path.
- Local users who prefer a guided interface can control page opening, capture, analysis, and export from the Streamlit application.
- Developers experimenting with Chinese conversational models can train the separate Seq2Seq chatbot on their own corpus and launch its visual chat module.
What are this agent's strengths and limitations?
- Provides an end-to-end workflow spanning page opening, flexible capture, batch extraction, review, and Excel generation.
- Offers full-screen, element-level, and manual-region capture modes, allowing users to limit model input to relevant order content.
- Separates the FastAPI backend from the Streamlit interface while exposing chat-driven controls and configuration options.
- Can operate against order portals that do not expose a direct export feature or documented data interface.
- Requires a local Chrome installation and a compatible ChromeDriver; version or path mismatches can prevent browser startup and capture.
- Extraction depends on an external model, API credentials, and network access, but the provider, pricing, and data-handling boundary are not documented.
- The stated Chrome requirement is inconsistent: one section requires version 140.0+, while troubleshooting discusses version 112.0+.
- Recognition quality depends on screenshot clarity and selection scope, so exported results require validation.
- No Docker deployment, authentication, permission isolation, production topology, or operational security guidance is provided.
How do you install or deploy this agent?
ShoppingAgent requires Python 3.8+, Google Chrome, and a matching ChromeDriver. The prerequisites specify Chrome 140.0 or later, while troubleshooting refers to compatibility with Chrome 112+, so the actual supported version should be verified before deployment. Clone the repository and run pip install -r requirements.txt from its root. Configure the server address and port, AI model parameters, API key, Chrome path, and driver path through the available settings or project configuration. The source does not provide a named model provider, an exact API-key variable, or a complete .env example. Start the backend with python -m app.main, then start the frontend in another terminal with streamlit run app/frontend.py. Open http://localhost:8501 if the browser does not launch automatically.
How do you use this agent?
Start both the FastAPI backend and Streamlit frontend. Enter an order-page address in the URL field and select the control that opens the page. Choose a full-screen, element-selection, or manual-region capture. For a manual region, first acquire the full-page screenshot, drag over the relevant order area, and select the action that captures the chosen region. After collecting images, run the batch order analysis, review the extracted records in the right-hand panel, and use the Excel export action to save them. Restart the application after changing model, API-key, server, browser, or driver settings.
FAQ
Can it connect directly to an order database or a commerce platform API?
Which model provider does ShoppingAgent require?
Where are screenshots and exported records stored?
screenshots/ and excels/ directories. It does not explain how an external model service retains or processes submitted image content.What should I check when Chrome or screenshot capture fails?
CHROME_PATH, confirm that Chrome and ChromeDriver versions match, wait for the page to finish loading, keep the browser window visible, and try another capture mode.