Show HN: Cerebellum – small model on the robot, big model for what it can't undo

Category: ai-ml

Tags: robotics, edge-inference, llm-routing, offline-first, fleet-management

Score: 6.5/10 (Innovation: 6, Technical: 7, Documentation: 8, Utility: 5)

Cerebellum is an open-source decision layer for robot fleets that routes each action by its reversibility: a small on-device model handles fast, safe decisions while NVIDIA Nemotron in the cloud approves irreversible or uncertain ones. It journals decisions locally in SQLite for offline resilience and runs one server/dashboard for the whole fleet. The routing-by-reversibility concept and NAT-friendly robot sync are a thoughtful combination, though the project is early-stage (3 stars) and tied to a hackathon demo.

Target audience: robotics engineers, edge AI developers, backend devs building physical-AI systems

Repository: https://github.com/cerebellum-labs/cerebellum · TypeScript · MIT · 3 stars

View on Hacker News