Skip to main content

Yugma Platform

Yugma is an AI-native 3D studio — a professional web-based scene editor where natural language is the primary interface. Users describe what they want ("build a sci-fi hangar with neon lighting"), and an AI Director composes the scene using structured tool calls against a live Three.js scene graph.

The product is NOT a 3D editor with AI bolted on. It is an AI-native scene graph that happens to render in 3D.

Platform Overview

ServiceURLPurpose
Studio Appyugmastudio-app.web.app3D scene editor with AI composer + real-time collaboration
Admin Paneladmin.yugma.studioUser management, provider config, usage stats
Documentationdocs.yugma.studioThis site
Landing Pageyugma.studioMarketing site

Tech Stack

LayerTechnology
FrontendReact 18, TypeScript, Vite
3D EngineThree.js via React Three Fiber + Drei
StateZustand (17 stores)
AIFrontier LLMs managed server-side — Gemini default (gemini-3.5-flash); Anthropic / Cerebras as server-configured overrides (no user-facing model picker — removed 2026-07-08)
3D GenerationMeshy.ai API (text-to-GLB)
AuthFirebase Auth (Google OAuth only)
DatabaseFirestore + Realtime Database
StorageFirebase Cloud Storage
FunctionsFirebase Cloud Functions v2 (Node 22)
HostingFirebase multi-site hosting

What Makes Yugma Different

  1. Agentic AI loop — The AI calls tools, sees results (including generated IDs), and chains up to 5 dependent operations per turn (default 3). It builds coherent scenes, not random objects.

  2. ~60 tokens/object (measured) — The custom YSL serialization format fits a 50-object scene in ~3,000 tokens, letting the AI maintain full scene awareness.

  3. Client-side spatial intelligence — Circle, grid, stack, spiral, and scatter patterns are computed with real trig, not hallucinated by the LLM.

  4. 16-phase roadmap — From scene editing (live) through digital twins, physics simulation, video reconstruction, and cinematic rendering (planned — earlier prototypes were removed in the 2026-07 cleanup).