Lawful magic over one world model
Magic is treated as a stateful engineering discipline. Spells compile into common state-transform logic so interactions emerge from the laws instead of bespoke pair scripting.
Research program
Zeta
Zeta is a preproduction tactics RPG and simulation program built around lawful magic, local LLM companions, and one rule-bound world model for humans, models, and deterministic agents.
Core pillars
What makes Zeta distinct.
Zeta matters on the site because it shows the longer arc of the company’s interests: not just models that answer questions, but systems where humans and model agents share a bounded world.
LLM-first party design
The design target is mixed human and model parties where local LLM companions act through structured intents under the same authoritative rules as everyone else.
Close-range mage play
The roadmap rejects artillery-only spellcasting. Posture, disruption, footing, visibility, and medium control are meant to matter as much as raw output.
Mastery over grind
Advancement is supposed to follow evidence of learning and adaptation, with anti-grind logic preventing trivial repetition from becoming the optimal path.
Prototype roadmap
Current build order.
The roadmap is explicit about sequencing: build the simulation substrate, validate the spell laws, prove close-range combat, and only then widen the content surface.
-
Phase 0 foundations: schemas, operators, spell compiler skeletons, and debug visualizations.
-
Phase 1 state sandbox: temperature, moisture, pressure, charge, opacity, structural integrity, momentum, and biological stress.
-
Phase 2 spell runtime: benchmark spells such as Cinder Line, Vapor Mantle, and Aegis Frame.
-
Phase 3 duel prototype: prove disruption, positioning, and close-range viability.
-
Phase 5 local companion: one useful, bounded local LLM party member.