OS for Cosmos

Overview

OS for Cosmos is an open-source AI research copilot for astronomy, still early-stage and built in public. It scans public survey data — Gaia, TESS, Kepler, JWST, Hubble, SDSS, LIGO — runs anomaly detection to surface objects that don't look like their statistical neighbors, and explains every flag with the evidence behind it instead of a bare confidence score. It's designed to run entirely on local, free tooling — no API keys, no cloud bill, no vendor to depend on. The roadmap runs from a validated detection loop (checked against Boyajian's Star, a known Kepler anomaly) through a local explanation layer and literature cross-referencing, toward a full researcher workstation called the Sky Explorer.

Highlights

• Open-source, local-first AI research copilot for astronomy — no API keys, no cloud bill, runs entirely on the researcher's own machine
• Anomaly detection pipeline scores objects against their statistical neighbors across public surveys (Gaia, TESS, Kepler, JWST, Hubble, SDSS, LIGO)
• Core detection loop validated against Boyajian's Star, a known Kepler light-curve anomaly, before trusting it on unlabeled data
• Every flag ships with the evidence behind it — features, comparisons and reasoning — instead of a bare confidence score
• Stack chosen for zero required cloud spend: Python pipeline, DuckDB for local storage, PyTorch for detection models, Ollama for local LLM explanations
• Roadmap: expand detection across more surveys, add an Ollama-based explanation layer, cross-reference arXiv/NASA ADS/Semantic Scholar, then build the Sky Explorer workstation UI

Screenshots