◈
Managed inference
Run supported world models through one clean API. No provisioning, environment setup, or serving stack to maintain.
WORLD MODEL INFRASTRUCTURE
Run world models.
Not GPU infrastructure.
Managed inference for emerging world models and physical AI. Submit jobs, run models, and retrieve outputs through a simple developer API.
Private preview in development
SCENE / 0047WORLD MODEL OUTPUT
▲
DEPTHSEGMENTATIONGENERATION
THE PLATFORM
World models, without the infrastructure overhead.
Build with the models you care about. ecru is shaping a focused execution layer for demanding physical-world workloads.
↗
Asynchronous execution
Submit long-running workloads, track their status, and retrieve completed outputs when they are ready.
□
Portable outputs
Take generated media and execution metadata into the tools and pipelines you already use.
MODEL EXPLORATION
Emerging models.
One execution layer.
We’re starting with open-weight physical-AI models and synthetic-data workflows, prioritised around real developer demand.
PHYSICAL AI
INFERENCE VALIDATED
NVIDIA Cosmos3 Nano Sim Transfer ↗
Depth-conditioned photorealistic video generation, successfully executed on NVIDIA H100 and RTX A6000 GPUs. Managed API access is being prepared for private preview.
INFERENCE VALIDATEDPLANNED API
Built for developers.
A straightforward workflow for long-running model inference.
ILLUSTRATIVE API — PLANNED INTERFACE
curl -X POST https://api.ecru.dev/v1/generations \
-H "Authorization: Bearer $ECRU_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "nvidia/cosmos3-nano-sim-transfer",
"input": { "depth_video_url": "https://example.com/depth.mp4" }
}'
→ 201 { "id": "gen_example123", "status": "queued" }
GET /v1/generations/gen_example123
→ 200 { "status": "completed", "output": { "video_url": "https://example.com/output.mp4" } }01Submit
02Monitor
03Retrieve
MODEL DEMONSTRATION
Real outputs from physical-AI inference.
Depth-conditioned photorealistic video generation using managed GPU infrastructure.
PRIVATE PREVIEW
What are
you building?
We’re working with early developers to shape the first supported models and workloads. Tell us what you’re trying to run.