A guide star for your
constellation of agents.

A living model of your world. GEB-1 learns from your agents and updates itself based on your goals. Coherence grows with the model. Control stays with you.

Request access → Read the paper

λ · 589.0 ㎚ Altitude · 92 km Rep rate · 13.5 kHz ● Guidance Active

Chapter 01 · GEB-1 Learns Constellation

A single reference everything else calibrates against.

GEB-1 is a world model trained online. It watches every agent in your Constellation, infers state and goal, and emits continuous guidance that bends the system toward what matters. Active inference - not another scheduler.

INFER

Posterior beliefs, live.

GEB-1 maintains a distribution over hidden state across every agent, updated each step from real traces.

PREDICT

Unified world model.

Simulates the Constellation forward, unifying your system's entire state into a cohesive predictive model.

LEARN

Active inference.

Over time, the agents in your Constellation learn the mapping between their local environment and your high-level goals.

OPTIMIZE

Free-energy descent.

Agents take action to align themselves with your goals based on the global world model. The collective system optimizes your goals over time.

Chapter 02 · Terminal Commands Constellation

The ground station.
Where you take the controls.

Sodium Terminal is where your team commands the system. Real-time Constellation view, guidance overrides, policy controls, and the complete log of every belief, action, and outcome your agents have ever produced.

— Observe

Live Constellation topology, beliefs, and guidance signals in one view.

— Command

Set goals, constraints, and overrides. GEB-1 re-plans in under a second.

— Audit

Every decision is traceable. Every override is signed.

A guide star for your constellation of agents.

A single reference everything else calibrates against.

The ground station. Where you take the controls.

A guide star for your
constellation of agents.

The ground station.
Where you take the controls.