get-physics-done — open-source agentic AI physicist (PSI)

Abstraction: Open-source agentic AI system for physics research

Key points:

• GPD (v1.2.2, Apache-2.0) by Physical Superintelligence PBC (PSI): an agentic AI physicist that installs physics-research commands into AI runtimes rather than being a standalone app.
• Supports five runtimes — Claude Code, Codex, Gemini CLI, GitHub Copilot CLI, OpenCode — installed via npx -y get-physics-done; needs Node 20+ and Python 3.11+.
• Four-stage workflow: Formulate → Plan → Execute → Verify (dimensional consistency, limiting cases, symmetry, conservation laws, numerical stability). Emits real artifacts: PROJECT.md, ROADMAP.md, .tex derivations, .py verification scripts.
• Hierarchy Project → Milestone → Phase → Plan → Task, with dependency "waves" for parallel execution; locks conventions across up to 18 physics fields; three capability tiers and model profiles (deep-theory, numerical, exploratory, review, paper-writing).
• Publication loop (write-paper → peer-review → respond-to-referees → arxiv-submission) and shared Wolfram integration; favors scientific rigor over agreeability ("a scalpel, not an autopilot"). Name is analogy to GSD (get-shit-done).

Connections: Get Physics Done · Physical Superintelligence · Claude Code · Wolfram · AI Agents · Agentic Coding · Scientific Research Automation

Source: https://github.com/psi-oss/get-physics-done