An open-source, continual-learning autonomous control system for advanced molten-salt microreactors — and, eventually, a complete reference design of one.
Prometheus stole fire from the gods and gave it to humanity. Promethea is the gift, not the thief — small, portable, and meant to be handed to whoever needs it.
Promethea is a public, end-to-end engineering project to design and simulate a community-scale molten-salt microreactor — and to build the AI-driven control system that makes it walk-away safe and autonomous.
It is being built by one person, in the open, on a workstation. Nothing here is, or ever will be, a real physical reactor. Everything here is software, simulation, and engineering documentation. The goal is not to build a reactor in a garage. The goal is to produce the artifacts that make the next generation of small-reactor builders 10× more productive, and to put a credible, reproducible design study on the internet for anyone to learn from, fork, or improve.
The grid is under stress from rising demand, aging hardware, and adversaries who have already pre-positioned inside critical infrastructure. Solar and storage solve part of the problem. Microreactors solve another part — dispatchable, energy-dense, weather-independent power small enough to live near the communities that need it.
Advanced reactor companies are tiny teams (often 20–50 engineers) and they are starved for: better open simulators, validated reference designs, modern controls research, and technically literate evangelists. A motivated independent contributor with the right tools can become a known quantity in this world in 12–18 months.
This is rooted in a conviction that good technology should serve the people who need it most, and that the asymmetry between those with electricity and those without will only grow in the AI era.
Working concept (v0 — subject to revision):
- Type: Chloride-salt, fast-spectrum, thorium-fueled microreactor
- Power: ~5 MWe / ~15 MWth (matches Westinghouse eVinci scale)
- Primary: NaCl–MgCl₂–ThCl₄ + HALEU starter (U-235 ~19.75%)
- Secondary: Heat-pipe array, eVinci-style (walk-away safe, no high-pressure water)
- Power conversion: Supercritical CO₂ Brayton cycle (~45% efficiency)
- Refueling interval: 8+ years
- Spectrum: Fast — no moderator, online actinide management possible
- Control: Continual-learning AI controller (hebbnet) running on embedded hardware, with classical PID/MPC fallback
This concept is positioned in a niche the major players have largely skipped: the intersection of chloride + fast + thorium + microreactor scale. Most thorium designs are fluoride/thermal (LFTR, Copenhagen Atomics, Seaborg). Most chloride fast reactors are uranium or waste-burning at utility scale (TerraPower MCFR, Moltex SSR-W, Elysium MCSFR). The combination at ~5 MWe is largely open ground in the published literature.
Phase 1 — Promethea Control (months 1–8). Reproduce the MSRE benchmark in OpenMC + Griffin. Build a digital twin and a promethea-gym environment. Implement PID + MPC baselines. Train a hebbnet continual-learning controller. Publish.
Phase 2 — Promethea Design (months 9–18). Lock the v0 design parameters. Iterate the neutronics in OpenMC until cold criticality, breeding ratio, and temperature coefficient are all in spec. Couple to OpenFOAM for thermal-hydraulics. Run LOFA / LOHS / RIA transient cases. Bolt the Phase 1 controller onto the Phase 2 design. Publish.
Phase 3 — Promethea System (months 19–30). Balance of plant, fuel cycle, deployment scenarios, cost-of-electricity estimate. End-to-end reference report. Companion site and documentary-style writeup.
See docs/ROADMAP.md for the detailed milestone breakdown.
This repo is the source of truth. Every design decision, simulation input deck, controller checkpoint, and write-up lives here. The aim is for any reactor physicist, ML researcher, or curious engineer to clone the repo, run the code on their own workstation, and reproduce every claim.
- Watch the repo for milestone releases
- Open issues with questions, corrections, or suggestions
- Pull requests welcome — see CONTRIBUTING.md
- Build log: posted to the project blog and YouTube channel (links coming)
You don't need a Linux box to reproduce Promethea results.
- GitHub Codespaces. Click
Code → Codespaces → Create codespace on main. The container builds OpenMC + the scientific stack automatically. Then in the terminal:bash scripts/fetch_xs.sh(one-time, ~4 GB),python scripts/hello_reactor.py(smoke test),python benchmarks/msre/run_criticality.py(full MSRE v0). Free tier covers it. - GitHub Actions. The
benchmark-msreworkflow runs the full MSRE criticality benchmark on every push tobenchmarks/msre/**, and can be triggered manually from the Actions tab with custom particle/batch counts. k-eff and PASS/REVIEW status are written to the run summary.
- This is not a real reactor. It is a simulation and engineering study.
- This is not an investment vehicle. There is no token, no company (yet), no fundraise.
- This is not an attempt to circumvent nuclear regulation. Any future physical work would go through the NRC like everyone else.
- This is not a claim of expertise. It is a public learning project that aims to produce real artifacts along the way.
- OpenMC (LANL et al.) — Monte Carlo neutron transport
- MOOSE / Griffin / BISON (INL, ANL) — multiphysics framework, reactor physics, fuel performance
- OpenFOAM — computational fluid dynamics
- OpenModelica — systems modeling
- hebbnet — neuromorphic on-device continual learning (github.com/KhaiB10/hebbnet)
- The ORNL MSRE archive — 60 years on, still the foundational dataset for molten salt reactors
- The openmsr community — github.com/openmsr
Code: MIT. Documents, designs, and figures: CC BY 4.0.
If Promethea is useful in academic or technical writing, please cite the
version you used. A machine-readable CITATION.cff is
included; GitHub renders it as a "Cite this repository" button at the
top right of the repo page.
BibTeX (v0.2.0):
@software{bui_promethea_2026,
author = {Bui, Khai},
title = {{Promethea v0.2.0: An open, CI-validated OpenMC
benchmark of the Molten Salt Reactor Experiment}},
year = {2026},
month = {5},
version = {0.2.0},
url = {https://github.com/KhaiB10/promethea/releases/tag/v0.2.0},
license = {MIT}
}For citations of specific results (k-effective values, sensitivity studies, primary-source audits) please also reference the underlying ORNL technical memoranda (TM-0728 and ORNL-4528) directly. A Zenodo DOI will be minted on the next tagged release.
Repository mirrors (read-only): GitLab · Codeberg — see docs/MIRRORS.md.
Khai · Wichita, KS · khaibustos10@gmail.com