Agents

Deploy a signal

Holders spend $luna to deploy a signal. A population of GPT-5.6 Luna agents, the fast execution tier, independently claims it and runs it in parallel. A bigger tier assigns more agents and higher queue priority: Crescent runs a cluster of 3 agents, Eclipse 8, Halo 16, and Max 32.

The three brands are the gpt-5.6 family and they form an escalation ladder. Luna is the population, fast and cheap, and handles almost everything. Terra is the council, called in to resolve ambiguity when the population splits. Sol is the oracle, rare and expensive, reserved for high-complexity disputes. The tiers mirror real effort modes, so the model is grounded in capabilities that actually shipped, not invented for the token.

What runs today: a population of parallel model calls on the gpt-5.6 family independently processes each signal, and the result settles when they agree. The /deploy page plays the ritual out, and POST /api/signal runs the real thing.

The lifecycle

A signal moves from deploy to settlement. The /deploy ritual plays it out as light: the signal ignites, agents come online as they claim it, and the field ripples cold when the epoch commits. Consensus and escalation, below, are how a result is decided.

Consensus and escalation

Consensus settles most signals: redundant agents run the same task and the result lands when enough of them agree. When they disagree, or confidence is low, the signal escalates. Luna handles roughly 70 percent of the work, Terra resolves about 20 percent, and Sol is the rare oracle for the last 10 percent. Escalation is meant to be rare, so most of the population's work resolves fast and cheap at the Luna tier.