CLI-first cloud GPU runtime

Serverless GPU infrafor agents withsub-5s cold boots.

Graphene gives humans and coding agents a CLI-first GPU runtime with production-equivalent CUDA, CAS/NVMe caching, and no SDK rewrite or Docker rebuild loop.

CLI install
Install the public Graphene CLI
brew install graphene-sh/tap/graphene
graphene version
graphene login
02/ CLI path

Init, add, run. The GPU path stays CLI-native.

A project locks its runtime, dependencies become cached artifacts, and the Python entrypoint starts on a live GPU.

01
Init
02
Add
03
Run
Terminal
l4 / python 3.11
03/ GPU lineup

T4 and L4 are live. Larger accelerators are queued next.

The public runtime catalog exposes the GPUs users can actually run today. The rest of the rate card stays visible so teams can plan for heavier agent and training workloads.

Supported today
T4 / L4

These are the GPUs exposed through the CLI runtime selector and backed by the current GCP deployment path.

graphene init --gpu l4
GPU
VRAM
Price
Status
NVIDIA T4
t4
16 GB
$0.013 / min
Available
NVIDIA L4
l4
24 GB
$0.021 / min
Available
NVIDIA V100
v100
16 GB
$0.083 / min
Coming soon
NVIDIA H100
h100
80 GB
$0.327 / min
Coming soon
NVIDIA B200
b200
180 GB
$0.600 / min
Coming soon
04/ No rewrite

Do not reshape your app for someone else's SDK.

Most GPU serverless products ask you to wrap functions, decorators, or app objects. Graphene starts from the command you already run.

Existing commands
graphene run -- python inference.py

Run an existing script on a cloud GPU without moving it into a new SDK shape.

graphene run -- uvicorn app:app

Expose the same service entrypoint you already use locally.

graphene run --train train.py

Submit longer GPU jobs while keeping CLI logs and status attached.

Shape mismatch removed
Local repo
python app.py
Cloud GPU
same command

The platform handles runtime selection, source sync, scheduling, and output streaming around your command instead of forcing a new application model.

05/ Cold start

Startup measured in seconds, not deployment cycles.

Graphene moves slow work out of the hot path. Runtime and dependency artifacts are prepared before a GPU run, then placed close to the selected node.

Time to user code
Graphene cached runseconds
SDK serverless cold starttens of seconds
VM provision + image pullminutes
What is already ready
01
Runtime

base + Python + CUDA

02
Deps

uv lock + installed tree

03
Code

small snapshot diff

04
Volume

model and data assets

sync path: resolve -> build -> pin -> cache
run path: snapshot code -> schedule -> execute
06/ Runtime parity

Stop debugging CUDA drift after deployment.

Graphene locks Python, CUDA, framework versions, and GPU class into a runtime profile. Local development invokes the same production-equivalent runtime that serves the job.

One runtime contract
Python
3.11
CUDA
12.x
Torch
pinned
GPU
T4 / L4
Dev and prod share the profile
01
Local CLI
graphene run
02
Runtime lock
sha256:profile
03
Production GPU
same contract
07/ CAS cache

No repeated Docker build loop on the GPU hot path.

Graphene treats heavy layers and fast-changing code differently. Runtime and dependency artifacts are content-addressed and reused; code moves as a small snapshot.

Traditional loop
edit01
docker build02
push image03
pull image04
test on GPU05
Graphene loop
Heavy artifacts
cache hit
+
Code snapshot
small diff
graphene run
08

Ship your first GPU run in minutes.