Auto-Analyst: A Chat-Based Foreman for Python Data Tools

What it does Auto-Analyst is a browser-based chat platform for analyzing datasets. It accepts CSV or Excel uploads, ingests a short text description of the dataset, and answers natural-language questions through a chat interface. A central planner routes queries to specialized agents — preprocessing, statistical analysis, machine learning, or visualization — each generating Python code with standard libraries like pandas, statsmodels, scikit-learn, and plotly. The project is open source under MIT and model-agnostic, requiring you to bring your own LLM API key.

The interesting bit The agents are implemented as dspy.Signature classes, which makes the framework essentially a well-structured orchestration layer rather than a reinvention of the underlying math. The planner can chain agents together for multi-step workflows, and a built-in retriever picks chart formats so you do not have to argue with plotly about whether your data wants a bar chart or a scatter plot.

Key highlights

• LLM-agnostic backend: works with OpenAI, Anthropic, DeepSeek, and others via your own API key.
• Modular agent system built on DSPy; custom agents for marketing, finance, or web APIs can be added by subclassing signatures.
• Built-in connectors for ad platforms (Google Ads, Meta, LinkedIn), CRMs (HubSpot, Salesforce), and SQL databases (Postgres, MySQL, Oracle, DuckDB).
• Includes a code editor to inspect and edit generated Python, plus an enterprise tier with usage dashboards and scheduled report regeneration.

Caveats

• Features like the analytics dashboard and daily scheduled reports are flagged as “Enterprise,” though the README does not clearly delineate what is open-source versus commercial.
• Multi-CSV and multi-sheet Excel analysis are on the short-term roadmap but not yet implemented.

Verdict Data analysts who want a conversational interface over familiar Python stacks should take a look; if you prefer writing notebooks by hand or need deep custom modeling, this will feel like polite middleware.