Repolode
Submit
plexe-ai

plexe-ai/plexe

Python tool that builds tabular machine learning models from a plain-language intent and a dataset using LLM-driven agents.

Open-source alternative to:DataRobotH2O Driverless AI

View on GitHubWebsite
Stars
2,642
Forks
263
Open issues
18
Last commit
8 months ago

What is it?

plexe lets you describe a machine learning model in plain language, point it at a tabular dataset, and get a trained model back. A multi-agent system handles the workflow: it analyzes the data, identifies the ML task, selects an evaluation metric, and iterates on candidate models to find a good one.

The project supports XGBoost, CatBoost, LightGBM, Keras, and PyTorch for tabular data, and can be run from the command line or as a Python function. Output is a self-contained model package with artifacts, inference code, input and output schemas, and evaluation reports, which does not depend on plexe at deployment time. LLM calls are routed through LiteLLM, so different agents can use different providers, though the project actively tests only OpenAI and Anthropic models.

plexe also ships Docker images with PySpark and Java preinstalled, a YAML configuration file, and a Streamlit dashboard for browsing experiment results and search trees.

Key features

  • 01Builds tabular models from a natural language intent and a dataset in Parquet, CSV, ORC, or Avro.
  • 02Uses a multi-agent workflow to analyze data, pick metrics, and iterate on candidate models.
  • 03Supports XGBoost, CatBoost, LightGBM, Keras, and PyTorch.
  • 04Exports a self-contained model package that can be deployed without plexe.
  • 05Routes LLM calls through LiteLLM to multiple providers, configurable per agent.
  • 06Includes Docker images with PySpark and a Streamlit experiment dashboard.

Use cases

  • Prototyping a tabular prediction model, such as churn or price prediction, without writing training code.
  • Generating a baseline model quickly to compare against hand-built pipelines.
  • Packaging a trained model as a standalone artifact for deployment in another environment.

Topics

#machine-learning#agentic-ai#llm#automl#tabular-data#mlops#agents#ai#ml#mlengineering

FAQ

What is plexe?
Python tool that builds tabular machine learning models from a plain-language intent and a dataset using LLM-driven agents. plexe lets you describe a machine learning model in plain language, point it at a tabular dataset, and get a trained model back. A multi-agent system handles the workflow: it analyzes the data, identifies the ML task, selects an evaluation metric, and iterates on candidate models to find a good one.
Is plexe free to use? What license does it have?
plexe is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is plexe still maintained?
No commits for 6+ months. The last commit to plexe was on Mar 6, 2026.
What can I use plexe for?
Prototyping a tabular prediction model, such as churn or price prediction, without writing training code. Generating a baseline model quickly to compare against hand-built pipelines. Packaging a trained model as a standalone artifact for deployment in another environment.
Where can I get plexe?
The source code and installation instructions are on GitHub: https://github.com/plexe-ai/plexe

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects