ADA launches MIT‑licensed open‑source AI data analyst

ADA released on July 20 2026 as an MIT‑licensed project by Sainesh Nakra. It accepts a CSV or Excel file, builds a Streamlit dashboard, runs pandas for aggregation, anomaly detection, and forecasting, and returns plain‑English answers. An optional LLM layer generates natural‑language explanations but only receives column names, so raw rows never leave the host machine [DevTo][GitHub].

What shipped

The initial release includes three components: (1) a pandas‑backed calculation engine that runs locally, (2) a Streamlit UI that visualizes metrics, anomalies, and forecasts, and (3) a toggleable LLM wrapper that operates without an API key and can run against any compatible local model. The repository provides a live demo at automated-data-analyst.streamlit.app and lists open issues for removing remaining LLM dependencies.

Why it matters

By keeping all numeric work in pandas, ADA guarantees deterministic, reproducible results, eliminating the “different number on re‑ask” problem common in pure LLM pipelines. Because the LLM only sees column headers, sensitive row data never traverses external APIs, aligning with GDPR‑style data‑locality requirements. The MIT license invites engineers to replace the optional LLM with domain‑specific models or add new analytics modules without licensing friction, making ADA a viable foundation for regulated‑industry pipelines.

ADA’s architecture flips the conventional LLM‑first mindset. Anchoring the workflow in pandas provides auditability and data residency needed for finance or healthcare workloads, even if it sacrifices the “one‑prompt‑does‑everything” convenience many startups tout.