Aurora

Aurora is glass-box quantitative intelligence — local, open, and cited. It's the verification layer for serious quantitative work: for humans analyzing hard data, and for AI systems that can't afford to hallucinate.

Every AI today invents numbers on quantitative claims — bigger models and RAG don't fix it. Aurora is the structurally different fix: it computes and verifies instead of predicting. Drop in a dataset (CSV, Parquet, JSON, XLSX) and it runs 24+ research-grade methods — Isolation Forest, robust z-score, Granger causality, HMM regimes, SINDy physics discovery, Gaussian processes, persistent homology, and more. Every finding is a structured object with a method, severity, threshold, and a citation linking to the exact knowledge-bank entry behind it — real sources like Newton, Granger, NIST, and NOAA. No invented numbers, no invented papers. A live "0 fabricated" chip is the contractual signal that every claim traces to a computation.

Cloud LLMs guess. Aurora computes.

Aurora has two faces sharing one engine. Copilot is the local studio for analysts, quants, scientists, and engineers — six analytical lenses, a spacetime system graph, and phase-space projection for exploring findings visually. Cortex is the verification layer AI builders call so their agents stop hallucinating math — via a Python SDK, an MCP server (works with Claude Desktop, Claude Code, Cursor, and custom agents), and Decision Contracts that fire webhooks or actions when findings match a condition. Every layer produces a portable, signable .aurora.json bundle with SHA-256 integrity and optional Ed25519 signing, so any result can be verified on another machine and audited for tampering.

Core principles: glass-box at every layer, local-first always (your data never leaves your machine — no telemetry, no phone-home), an honesty rule that renders uncertain findings as uncertain and discloses any skipped or sampled methods, and fully open source under Apache 2.0 — inspectable, forkable, free to use commercially.

A typical run finishes in about 14 seconds on consumer hardware, fully local after the initial knowledge-bank download. Built in the open by one person, with an honest changelog of what's solid and what's still rough.

GitHub: github.com/FantasyLab-ai/aurora · Site: fantasylab.ai