# Sovereign Autonomy

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Othniel Obasi](https://cerebralvalley.ai/u/othnielobasi)
- **GitHub:** https://github.com/othnielObasi/sovereign-autonomy
- **Demo video:** https://youtu.be/Kak8teTVHSA
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4304

Project Description

Sovereign Autonomy is the command and trust layer for robot fleets. Robot fleets of carriers and humanoids now work beside people in pharma cleanrooms, hospitals and warehouses. The AI that plans their moves can be wrong, confused or manipulated, and in a cleanroom one bad move means an injury, a lost batch or a failed audit.

Sovereign Autonomy coordinates robot missions, governs execution, manages approved fleet knowledge and preserves evidence of what happened.

Plan: An operator assigns a mission, for example "carry the QC samples to packaging". Meta's Muse Spark plans the route from the site map and the robot's own camera.
Govern: Every proposed move goes through a safety gate that checks human proximity, speed, clearance, path and model confidence. Only an approved move is signed into a single-use permit, and the robot's controller verifies that permit before it moves.
Escalate: When a worker is too close, the move is refused and the robot stops or replans. When the system is unsure, it asks a human; once an operator approves, a fresh permit is issued.
Stop: Stops never depend on the AI. A button or a spoken command (Muse Voice Transcribe) halts one robot, one mission or the whole fleet; authority is revoked in milliseconds. Only a signed-in operator can release a stop.
Learn as a fleet: What one robot learns becomes a lesson only when two robots corroborate the evidence and an operator approves it. The lesson is then released only to robots it fits: carriers receive it, and the humanoid, with a different body, does not. Lessons expire after 24 hours.
Prove: Every decision leaves a signed, verifiable record, linked in a Memgraph evidence graph and traced with LangSmith and Prometheus.
Under the hood: NetworkX routing, canary tripwires for prompt injection, and checkpoints for safe recovery.

Live: in production on Nebius at https://204-12-178-126.sslip.io (HTTPS, Postgres, signed backups).

What was built when: A policy gate, decision receipts, a warehouse twin and Tavily site intelligence came from earlier work. Everything else, including all of the Meta integration, permits, kill switches, fleet learning, the mixed fleet, the pharma site and the deployment, was built during this hackathon.

First market: pharmaceutical manufacturing, then hospitals and warehouses.

## More from Meta Global AI Developer Hackathon

- [Nativa](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4301)
- [HomeSync](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4302)
- [XamKids](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4303)
- [Knight Riders](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4305)
- [Sentinel V4](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4306)
- [REALMFORGE LIVE](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4307)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4304. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt