MultiVis-Agent: A Multi-Agent Framework with Logic Rules for Reliable and Comprehensive Cross-Modal Data Visualization

MultiVis-Agent is a multi-agent framework with explicit logic rules for reliable, comprehensive cross-modal data visualization, supporting inputs such as natural language, code, and images. This repository contains both the research artifact for the SIGMOD 2026 paper and the interactive demo artifact presented at SIGMOD 2026 Demo. It also provides a web-based demonstration and video presentation of MultiVis-Agent, MultiVis-Bench, a benchmark for text-to-vis and visualization modification tasks, and an automatic metric suite for visualization quality.

Publications

• SIGMOD 2026 (Research Track): MultiVis-Agent: A Multi-Agent Framework with Logic Rules for Reliable and Comprehensive Cross-Modal Data Visualization
• SIGMOD 2026 (Demonstration Track): Demonstration of MultiVis-Agent: Interactively Generating Reliable Visualizations via Logic-Rule Agents

Demonstration

Demonstration_of_MultiVis-Agent.mp4

Repository Overview

• MultiVis-Agent/: Core multi-agent system (coordinator, tool manager, config, database/query and validation agents).
• MultiVis-Agent_demo/: Front-end of multi-agent system.
• MultiVis-Bench/: Benchmark datasets and reference implementations for text-to-vis and vis-modify tasks.
• metric/: Visualization evaluation metrics used in the paper.
• run_system.py: Example entry script to run the multi-agent visualization system on a sample.
• run_metric.py: Script to compute metrics over saved results.

Installation

• Set up environment

git clone https://github.com/Jinwei-Lu/MultiVis.git
cd MultiVis
pip install -r requirements.txt

• Configure LLM APIs

Edit MultiVis-Agent/utils/Config.py, MultiVis-Agent_demo/vis_system/utils/Config.py and replace the placeholder "xxx" values in MODEL_CONFIGS with your own API keys and endpoints.

Quick Start

• Run a sample visualization generation case

python run_system.py

The script creates log and temporary folders and runs a sample item through the CoordinatorAgent, logging intermediate steps and final visualization code.

• Run Front-end of multi-agent system

cd MultiVis-Agent_demo
python app.py

The script starts the frontend on a local port and users can interact with multi-agent system on the web page.

[Update] The UI has been optimized. The demo is kept as source code under MultiVis-Agent_demo/; install MultiVis-Agent_demo/requirements.txt and run python app.py from that directory to launch the frontend locally.

• Evaluate generated results

Organize your results under ./results/{method_type}/{model_type}/{data_type}/results.json following the structure in run_metric.py, then run:

python run_metric.py

This produces metric.json, wrong_results.json, and correct_results.json for each data type.

Benchmark Data

MultiVis-Bench provides benchmark files:

• text2vis.json, vis_modify.json, text2vis_with_img.json, text2vis_with_code.json under MultiVis-Bench/.
• database/ and img/ subdirectories with databases and images used in benchmark tasks.
• For the databases in MultiVis-Bench/database/, download the Spider dataset from Google Drive and extract all database files into the MultiVis-Bench/database/ directory.

License

• Source code in this repository is released under the MIT License. This includes the framework under MultiVis-Agent/, metric scripts, top-level runnable scripts, and Python reference implementations under MultiVis-Bench/code/ and MultiVis-Bench/vis_modify/. See LICENSE.
• Benchmark data and annotations distributed with this repository are released under CC BY 4.0. This includes the JSON benchmark files under MultiVis-Bench/ and other benchmark assets distributed with the repository unless otherwise noted. See LICENSE-DATA.
• Third-party resources that must be downloaded separately, including the Spider databases placed under MultiVis-Bench/database/, are not covered by the licenses above and remain subject to their original licenses and terms.

Citation

If you find this project useful in your research, please cite:

@article{10.1145/3786670,
author = {Lu, Jinwei and Song, Yuanfeng and Zhang, Chen and Wong, Raymond Chi-Wing},
title = {MultiVis-Agent: A Multi-Agent Framework with Logic Rules for Reliable and Comprehensive Cross-Modal Data Visualization},
year = {2026},
issue_date = {February 2026},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
volume = {4},
number = {1},
url = {https://doi.org/10.1145/3786670},
doi = {10.1145/3786670},
abstract = {Real-world visualization tasks involve complex, multi-modal requirements that extend beyond simple text-to-chart generation, requiring reference images, code examples, and iterative refinement. Current systems exhibit fundamental limitations: single-modality input, one-shot generation, and rigid workflows. While LLM-based approaches show potential for these complex requirements, they introduce reliability challenges including catastrophic failures and infinite loop susceptibility. To address this gap, we propose MultiVis-Agent, a logic rule-enhanced multi-agent framework for reliable multi-modal and multi-scenario visualization generation. Our approach introduces a four-layer logic rule framework that provides mathematical guarantees for system reliability while maintaining flexibility. Unlike traditional rule-based systems, our logic rules are mathematical constraints that guide LLM reasoning rather than replacing it. We formalize the MultiVis task spanning four scenarios from basic generation to iterative refinement, and develop MultiVis-Bench, a benchmark with over 1,000 cases for multi-modal visualization evaluation. Extensive experiments demonstrate that our approach achieves 75.63\% visualization score on challenging tasks, significantly outperforming baselines (57.54-62.79\%), with task completion rates of 99.58\% and code execution success rates of 94.56\% (vs. 74.48\% and 65.10\% without logic rules), successfully addressing both complexity and reliability challenges in automated visualization generation.},
journal = {Proc. ACM Manag. Data},
month = apr,
articleno = {56},
numpages = {25},
keywords = {multi-agent framework, natural language to sql, data visualization, database applications, automated code generation}
}

@inproceedings{lu2026demonstration,
  title={Demonstration of MultiVis-Agent: Interactively Generating Reliable Visualizations via Logic-Rule Agents},
  author={Lu, Jinwei and Lu, Jiawei and Zhang, Chen and Song, Yuanfeng and Wong, Raymond Chi-Wing},
  booktitle={Companion of the 2026 International Conference on Management of Data},
  year={2026}
}