SERVERLESS
Deploy. Ship,
Push AI code from a template or your own repo to run serverless GPU workloads. We handle containers, scaling, and cold starts — you pay only for the seconds your GPU is actually running.
per hour billing
0 → N auto-scaling
100s of templates
HOW IT WORKS
Serverless GPU from template to endpoint in
Choose a template
Deploy
Call your endpoint
Scale automatically
BUILT FOR VARIABLE WORKLOADS
Built for variable,
Per-second billing
Pay only while your function is executing.
Automatic scale-to-zero
No traffic, no cost — serverless GPU workers spin down when idle.
Burst without planning
Scale to many concurrent workers under load, with no capacity planning needed.
WHO IT’S FOR
Built for not managing servers
Product teams shipping inference APIs
Image generation, chat, transcription — features with unpredictable traffic on serverless GPU inference.
Developers who don't want ops
Image generation, chat, transcription — features with unpredictable traffic.
Fast prototyping
Start from a template before committing to a dedicated GPU Cloud instance.
SERVERLESS VS GPU CLOUD
Which one do i need?
Both live under the same account — move between them anytime.
GPU CLOUD
Raw GPU power,
From single A100s to multi-node H100 clusters — spin up bare-metal on-demand GPU capacity in under a minute with no scheduling queues.
GPU CLOUD
Git push
Connect your repository once and get fully automated build, test, and deployment pipelines. Zero-downtime blue/green deploys on every commit.
Connect you repo
One-click GitHub / GitLab integration. We listen to push events on any branch you configure.
Automatic container build
We detect Python, pin CUDA, cache your requirements.txt, and produce a reproducible image in seconds.
Canary → production
New builds get 10% of traffic first. If health checks pass, traffic shifts 100% with no downtime.
Instant rollback
Every deploy is versioned. Roll back to any previous build in one click or one CLI command.
SUPPORTED RUNTIMES
Bring your Stack.
Any Python framework, any model hub, any GPU. We auto-detect
and configure the right CUDA base image.
PyTorch
2.3 - CUDA 12.4
HuggingFace
Transformers 4.x
vLLM
0.5 - PagedAttn
TensorRT
10.x - FP8
Diffusers
SDXL - Flux
llama.cpp
GGUF - GGML
JAX
0.4 - XLA
AudioCraft
MusicGen - MACs
OpenMM
Protein - Mol
Custom
Any CUDA image
FAQ’S
Serverless GPU FAQ's
What is serverless GPU inference?
Serverless GPU inference lets you deploy AI functions that automatically scale to zero when idle and scale up under load, billing you only for the seconds your GPU is actively running — no idle capacity, no manual provisioning.
How fast does a serverless GPU function scale from zero?
[Insert real measured cold-start benchmark here once available — mean/median seconds from a scaled-to-zero function to first response, by model size.]
How is Serverless different from GPU Cloud?
Serverless deploys event-driven functions that scale to zero and bill per second — best for variable, unpredictable traffic. GPU Cloud gives you a dedicated, always-on instance with full SSH access, billed per hour — best for long-running training and fine-tuning.
Can I deploy my own Docker container?
Yes — deploy any custom container, or choose from 100s of pre-built templates for PyTorch, HuggingFace, vLLM, TensorRT, Diffusers, and more.
What happens when I push code to my repository?
Every push triggers an automated build, a canary deploy to 10% of traffic, health checks, and a zero-downtime rollout to 100% traffic if checks pass — with instant rollback to any previous version if needed.
What frameworks and runtimes are supported?
PyTorch, HuggingFace Transformers, vLLM, TensorRT, Diffusers, llama.cpp, JAX, AudioCraft, OpenMM, and any custom CUDA image.
