whereismyai
Cerebellum
Cerebellum is an open-source decision layer for robot fleets that routes tasks between local models and the cloud.
Cerebellum is an open-source decision layer designed for robot fleets to manage tasks between local device models and cloud models. It routes every decision based on predefined cost rules for mistakes, supporting reflex actions, reversible actions, and irreversible actions. The system uses NVIDIA Nemotron for reviewing unsure calls and handling irreversible actions that require cloud intelligence. It includes a Python agent library running on the robot with SQLite, a Go cloud server with PostgreSQL, and a React live web dashboard. The software keeps robots operating during network drops by journaling decisions locally and syncing them when reconnected.
Cerebellum at a glance
Key features
- Route decisions between local robot models and cloud models
- Handle reflex, reversible, and irreversible action classes
- Journal decisions locally with SQLite during network drops
- Sync decision logs and queues once the network reconnects
- Review unsure robot decisions asynchronously in the background
- Monitor fleet activity and costs using a web dashboard
Use it for
- Sort items with a robot camera and local detector
- Manage robot fleet decisions during network outages
- Review unsure robot actions with cloud models
- Audit robot actions through a centralized web dashboard
Alternatives to Cerebellum
- OpenSquilla — Maximizes AI intelligence density while minimizing token costs through smart routing.
- Bot Reborn — Runs directly in mobile Chrome over WebUSB with no server in between
- flightdeck — Monitor and control production and coding agents with real-time token budgets and kill switches.
- Robot Heads — Visualizes agent states with customizable robot head components.
- OpenHarness — Manages Claude Code, Codex, and Hermes from a single command center.
- Rakazo — Complete ownership of data, models, and execution environment.
- Agent Executor — Declarative runtime for scaling stateful agent workloads with isolation.
- Caura — Governed memory providing persistent knowledge sharing for AI agent fleets.
- FastGPT — It uses hybrid retrieval and auto-cleansed data to prevent hallucinations.
- Rax — Exploration compiles into code that runs up to 241 times faster.
- OpenBot — Runs entirely on your own infrastructure with strict document permissions and custom.
- NemoClaw — Provides managed security, networking, and lifecycle operations for AI agents.
All Cerebellum alternatives