Gaming AGI Open SDK

Deterministic by construction

Resolve simultaneous movement, consequence cascades, transport, projectiles, gates, triggers, rays, and scoring with replayable outcomes.

AI-native, not AI-attached

Expose concrete legal actions, deterministic seeds, transcripts, batch evaluation, model drivers, and MCP-capable CLI launchers from one environment contract.

Product-neutral mechanisms

The SDK owns how reusable mechanics behave. Your product retains characters, levels, objectives, tuning, authentication, and presentation.

Our mission

Make interactive games a shared, inspectable proving ground for human and machine intelligence—where both act through the same rules, face the same consequences, and can be compared through reproducible play rather than persuasive outputs alone.

Games measure behavior across a sequence of consequential choices. Grids keep that behavior structured and inspectable. Simultaneous turns remove reaction speed and request order as accidental advantages while testing prediction, coordination, and adaptation between actors.

One reducer, every way to play

The same injected deterministic reducer can power local play, hosted sessions, solvers, replay checks, model agents, CLI agents, and evaluation runs. There is no second, approximate implementation just for automation.

text

Product reducer + authored content
              |
              v
   GAOS deterministic engine
       /       |        \
  renderer   solver   AgentEnvironment
                         |
                 tools / drivers / CLIs

One agent turn

State	Legal actions	Agent chooses	Deterministic result
`position: 1` `status: playing` `actionsUsed: 1`	`{ id: 'advance' }` `{ id: 'jump', index: 2 }`	`{ id: 'jump', index: 2 }`	`position: 3` → won `reward: +3` · `totalReward: 3` · 3★

State

Legal actions

Agent chooses

Deterministic result

position: 1
status: playing
actionsUsed: 1

{ id: 'advance' }
{ id: 'jump', index: 2 }

{ id: 'jump', index: 2 }

position: 3 → won
reward: +3 · totalReward: 3 · 3★

AgentEnvironment exposes the product state and concrete legal actions, validates the agent's choice, applies the injected reducer once, and returns the scoring result with transcript-ready metrics.

Built with GAOS

Zonoid

The first game built with the GAOS Turn-Based Grid Toolkit.

Zonoid is a strategy game for humans and AI agents, built around prediction, planning, and judgment.

GAOS provides the reusable toolkit; Zonoid's product content stays separate from the SDK.

Visit Zonoid →

A deliberate ownership boundary

SDK owns

Reusable algorithms, deterministic ordering, settlement, protocol primitives, replay, scoring behavior, agent lifecycle, and integration contracts.

Your product owns

Characters, abilities, authored levels, game modes, objectives, thresholds, world tokens, prompts, hosting policy, seasons, and presentation.

The rule is simple: the SDK defines how a reusable mechanism behaves. The product decides where it is used and what it means.