Built during Artemis II — humanity's return to the Moon, April 2026
Space missions are becoming autonomous because physics forces them to. A signal to Mars takes up to 24 minutes one way — by the time a mission controller sees a problem and responds, an AI system has already been acting alone for nearly an hour. Human-in-the-loop control is physically impossible beyond the Moon.
The bottleneck isn't compute. It's trust.
Today, mission constraints are encoded as brittle hardcoded rules written by engineers — inaccessible to the scientists who understand mission intent, impossible to audit, and never systematically stress-tested before launch. When an autonomous system makes a consequential decision in deep space, nobody has a clear record of what it was supposed to do, or whether the rules it was given actually captured that intent.
AXIOM is a constraint specification and stress-testing interface for autonomous space systems. It lets mission planners write what they intend their system to do in plain language — then uses AI to structure those constraints, surface ambiguities, and act as an adversarial agent finding every edge case and loophole before the spacecraft leaves the ground.
The output is a validated, human-readable Mission Constraint Document — the governance artifact that stands between mission intent and autonomous action.
Built on Constitutional AI principles. Designed for the age of deep space autonomy.
Stage 1 — Mission Initialization Enter your mission name, type, and primary objective. Claude returns a structured Mission Context Card with inferred key constraints and operational environment.
Stage 2 — Constraint Authoring Write mission constraints in plain English. Claude parses each one, flags ambiguities, and assigns a priority tier — Safety, Power, Science, or Ops. The result is a living constraint hierarchy.
Stage 3 — Adversarial Stress Test Claude switches into adversarial mode — acting as an autonomous agent trying to find loopholes. It generates edge case scenarios and evaluates each against your constraint set, surfacing conflicts (two constraints contradict) and gaps (no constraint applies).
- React + Vite
- Tailwind CSS
- Anthropic Claude API (Sonnet 4.6)
- Vercel
git clone https://github.com/zainabmakhdum/axiom-space.git
cd axiom-space
npm installCreate a .env file:
VITE_ANTHROPIC_API_KEY=your_key_here
npm run devOpen localhost:5173
The hero image is NASA photo art002e009289 — the lunar surface photographed by the Artemis II crew on April 6, 2026, three minutes before Orion passed behind the Moon and lost contact with Earth for 40 minutes. NASA imagery is public domain.
AXIOM is a proof of concept. Stress test scenarios are generated by Claude based on logical analysis of constraint sets — not verified by certified mission planning tools. Designed to augment expert judgment, not replace it.