Private beta — game studios only

Send a prompt.
The forge routes it.

No model menus. No parameter tuning. No engine-specific prompt engineering. Z-Forge analyzes your request intent in real-time and dynamically routes it through the optimal domain-adapted model. One endpoint. One key. Zero decisions.

How it works
20+
Adapted model variants
<200ms
Routing + TTFT
0
Models you need to pick

Watch the forge route in real-time

Type a prompt or pick an example. See how Z-Forge classifies intent, injects context, and selects the optimal model — all in real-time.

One API call. Five-stage pipeline.

Every request passes through a domain-specific optimization pipeline before it reaches a model. You never see the complexity. You just get the best possible output.

01

Intent Classification

A lightweight classifier analyzes your prompt in real-time to determine the task domain — shader codegen, NPC dialogue, procedural generation, asset analysis, debugging, or architecture reasoning. No labels required.

02

Context Injection

Engine-specific schema graphs are automatically injected into the model context window. Unity ECS architecture trees, Unreal Gameplay Ability System chains, Godot node hierarchies — all handled transparently.

03

Model Selection

The forge routes to the optimal adapted model based on intent classification and context type. Claude Opus 4.8 for complex architecture reasoning, GPT-5.5 for real-time dialogue, DeepSeek V4 for cost-optimized shader codegen — the forge decides, not you.

04

Token Optimization

Domain-specific tokenizer re-sampling is applied based on the detected language and engine target. HLSL, GLSL, C#, C++, GDScript — each gets its own token distribution profile for maximum effective density.

05

Unified Response

The forge returns a single, coherent response through your standard OpenAI-compatible endpoint. No model names. No routing metadata. Just the output — optimized, adapted, and ready to use.

Stop choosing models. Start shipping games.

Without Z-Forge

  • Manually evaluate 20+ models for each task type
  • Write and maintain engine-specific system prompts
  • Switch API providers and keys when needs change
  • Manage multiple billing relationships
  • Tune parameters per model — temperature, top_p, context limits
  • Handle model deprecations and migrations manually

With Z-Forge

  • Send a prompt. Get the best possible output. That's it.
  • Engine context injected automatically — no prompt engineering
  • One API key, one endpoint, one bill
  • Optimal model selected per-request in real-time
  • Domain-adapted tokenization for shader, C#, C++, GDScript
  • New models added to the fleet automatically — no migration needed

We don't collect. We don't store. We don't train.

No telemetry. No prompt logging. No model training on your inputs or outputs. Your requests are processed in-memory and discarded immediately after response delivery. We retain only what's required to operate your subscription — an email and a billing relationship. Everything else is transient by design.

0
Prompts logged
0
Data retained
0
Models trained on you

Join the forge

Enter your access code to provision an account.

The engines behind the routing layer.

Z-Forge wraps frontier LLMs in a game-specific optimization pipeline. These are the systems that make automatic routing possible.

Intent Classification Engine

Real-time prompt analysis using a fine-tuned lightweight classifier trained on 2M+ game development queries. Detects task domain — shader codegen, NPC dialogue, procedural generation, debugging — with 94.3% accuracy at sub-10ms latency. No user-side labeling required.

Shader-Aware Token Re-sampling

Custom tokenizer profiles trained on HLSL, GLSL, MSL, and Shader Graph IR representations. 40%+ effective token density improvement for GPU pipeline code generation, with automatic bind-point semantic preservation and descriptor set layout inference.

Engine Schema Graph Injection

Unity ECS, Unreal Gameplay Ability System, and Godot 4 node DAG topologies are mapped to specialized adapter layers. Schema context is injected pre-attention, reducing cross-engine hallucination by up to 87% without any user-side system prompt configuration.

Procedural Generation Presets

Built-in task templates for Wave Function Collapse constraint propagation, noise field parameter space exploration, and automatic LOD hierarchy planning. The forge detects procedural generation intent and applies the appropriate generation graph automatically.

Runtime Dialogue Latency Tier

Dedicated low-latency inference channel with sub-200ms time-to-first-token. Persistent conversational state across NPC dialogue trees with branching context retention. The forge automatically detects real-time dialogue requests and routes them through this tier.

Asset Pipeline Context Retention

Extended context window with priority retention for asset pipeline metadata — UV layout descriptors, skeletal hierarchy hashes, and material property graphs persist across multi-turn iteration sessions. Zero context drift during long-form asset refinement.

20+ adapted models. You never pick one.

The forge routes your request to the optimal model from a fleet of domain-adapted frontier LLMs. New models are added to the fleet continuously — you get the upgrades automatically, no migration needed.

Claude Opus 4.8

Anthropic

Adapted
ZF-Opus-4.8-Game

ECS architecture reasoning, long-context agentic codegen

GPT-5.5

OpenAI

Adapted
ZF-GPT-5.5-Game

Real-time NPC dialogue, narrative branching & tool use

Gemini 3.1 Pro

Google

Adapted
ZF-Gemini-3.1-Game

Multimodal texture analysis, long-context asset Q&A

Grok 4.5

xAI

Adapted
ZF-Grok-4.5-Game

Real-time reasoning & procedural generation pipelines

DeepSeek V4

DeepSeek

Adapted
ZF-DeepSeek-V4-Game

On-prem inference, shader codegen & cost-optimized routing

Llama 5

Meta

Adapted
ZF-Llama-5-Game

Open-weight fine-tuning & custom engine adaptation

Mistral Large 3

Mistral

Adapted
ZF-Mistral-L3-Game

Low-latency European inference & GDPR-compliant pipelines

GLM-5

Zhipu

Adapted
ZF-GLM-5-Game

Multilingual game narrative & localization generation

+ 12 more adapted variants in active rotation · New models added monthly

One endpoint. Zero config.

OpenAI-compatible SDK. Install the Z-Forge package and start coding. No base URL exposed. No model parameter. No configuration files. The forge handles routing, authentication, and optimization internally. Works with Unity, Unreal, Godot, and any engine with HTTP support.

// Install the Z-Forge SDK
// Docs: docs.z-forge.ai/quickstart

using ZForge;

var client = new ForgeClient("sk-zf-...");

// Send a request — the forge auto-routes
var response = await client.CompleteAsync(
    "Generate a HLSL fragment shader for "
    + "stylized water with depth-based opacity"
);

// That's it. No base URL. No model selection.
// The forge handles routing internally.

Full documentation available in your dashboard

Quick start guides for Unity, Unreal, Godot & custom engines

Unity
Unreal Engine
Godot 4
Custom Engine

Ready to forge?

Private beta. Game studios and indie developers only. Invite-based access.