Pipelines · Agents · Retrieval · Semantic Engineering of Context
An open, forkable toolkit for grounding AI agents in real astronomy.
⚠️ Placeholder scaffold — early days. The structure and the contribution model are here first; the tools grow with the community. Expect this to change.
AI models are fluent and, on the specifics of your science, often quietly wrong. Grounding fixes that — put the right facts, data, and methods in front of the model at answer time, from a source you curate and can read.
PARSEC will be a collection of small, dependency-light MCP tools and patterns that do exactly that for astronomy: retrieval over your own knowledge base, survey / catalog access, pipeline and reproduction helpers — each readable in a sitting.
Science gets better when we work together. Fork it, add the tool your group needs, send it back. No API keys, no vendor lock. That's the whole point.
Want to see grounding change a real answer first? The hands-on companion is RAG over a knowledge base in ~90 lines of stdlib Python, no account, no cloud:
→ westoverlabs/parsec-rag-demo
That's the on-ramp. This repo is where the reusable tools will live and grow.
Read AGENTS.md — the shape and standards every tool follows
(applies to AI-agent contributors too). Then fork, add your piece, open a PR
describing what you verified by actually running it. Nothing is too small.
MIT — see LICENSE. Built by Westover Labs for the LSST / Rubin
community and anyone doing science with agents.