Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.
Engine bridges: Unity, Godot, and Unreal
A glTF viewer answers "does the mesh load." A game engine's import step asks a different question: can this ship in my project, with my pipeline, and drop into my scene with one click? The Forge API already generates the mesh; these three packages are the one-click part, one per engine, each calling the exact same public Forge API a browser would.
- Contract:
specs/ENGINE_BRIDGE.md, the shared auth/catalog/generate/poll/download/import contract all three bridges implement independently. - Unity:
packages/unity-bridge, a UPM package using glTFast to import. - Godot:
packages/godot-bridge, an editor addon using Godot's own nativeGLTFDocumentglTF importer. - Unreal:
packages/unreal-bridge, a plugin using Unreal's own Interchange glTF pipeline.
No bridge parses glTF itself. Each one is an HTTP client plus a thin adapter
that hands the downloaded .glb to the engine's own first-party importer,
the same split packages/blender-mcp and the Python Forge client vendored into the Blender/ComfyUI/Modly plugins already use.
Why one spec, three implementations
Unity (C#), Godot (GDScript), and Unreal (C++) have nothing in common at the
language level, but the API they're calling is one surface:
POST /api/forge to submit, GET /api/forge?job=<id> to poll, and a plain HTTPS GET to download the
finished .glb. Writing the contract down once in
specs/ENGINE_BRIDGE.md
means a fourth engine (or a bug report against one of the three) has a
single source of truth to check against, instead of three READMEs that can
quietly drift.
Every bridge exposes the same four import settings from the spec under the same names, mapped onto whatever each engine calls its own equivalent:
| Setting | Unity | Godot | Unreal |
|---|---|---|---|
generate_colliders |
MeshCollider (convex) |
ConvexPolygonShape3D |
Bounding-box UBodySetup |
import_materials |
glTFast materials, or a flat Standard |
glTF PBR materials, or stripped | Interchange materials, or UMaterial::GetDefaultMaterial |
scale |
Applied to the imported root's localScale |
Applied to the imported root's scale |
Applied by the caller at the actor level |
destination |
Scene root / parent transform | Scene tree parent | Content-browser path (/Game/...) |
Auth and routing
Generation is anonymous by default: each bridge generates a random
x-forge-client handle per client instance, the same free-first routing
and lane health every other Forge caller gets (see the pipeline
doc
for the backend grid and circuit breakers). Set a provider key on any client (ProviderKey in
Unity/Unreal, provider_key in Godot) to use your own BYOK geometry
provider instead of the free-tier lane.
Try it
- Unity: install via Package Manager git URL, then **Window > three.ws
Forge Bridge**. See
packages/unity-bridge/README.md. - Godot: copy the addon into
addons/, enable it in Project Settings, use the three.ws Forge dock. Seepackages/godot-bridge/README.md. - Unreal: copy the plugin into
Plugins/, enable Interchange glTF import, use Window > three.ws Forge Bridge. Seepackages/unreal-bridge/README.md.
Related: the Forge pipeline for how a submission is routed and served, Engine QA for grading whether an imported mesh will actually run at 60fps once it's in the engine.