v0.2 live — stage pipelines & secret vault

Serverless GPUs.Deploy in seconds.

One decorator. One command. A100s and H100s — auto-scaling from zero, billed only for what ran.

1.2s

avg deploy

idle cost

app.py

import gworker_client as gw
 
app = gw.App("transcribe")
 
@app.task(gpu=gw.Gpu("A100"), memory=8192)
async def run(audio_url: str) -> dict:
    import whisper
    model = whisper.load_model("base")
    return {"text": model.transcribe(audio_url)["text"]}

↳ Deployed in 1.2 s · A100 · US-east

Works with your stack

From idea to GPU in three steps

No infrastructure YAML. No provider accounts to wire up. Just Python.

Write

Decorate any async function with @app.task and specify your GPU. That's your entire deployment spec.

Push

Run gworker deploy. The CLI packages, uploads, and registers your worker in under 3 seconds.

Scale

Workers auto-scale from zero to thousands. You're billed per second of actual GPU time.

Everything you need. Nothing you don't.

Write once

One Python SDK with @app.task. No backend lock-in, no per-provider rewrites.

Ship anywhere

gworker picks the right GPU backend for you. You never wire it up.

Per-second billing

Transparent per-second GPU billing across US, EU, and Asia. No surprises.

Stage pipeline

Promote dev → uat → prod with one command. Logs stream to your terminal.

Cold start < 1s

Pre-warmed pool keeps your workers ready. No idle time means no waiting.

Secret vault

AES-GCM encrypted environment variables. Scoped per app, per stage.

Interactive Workloads

Explore typical serverless GPU workloads deployed on gworker. Change workloads to see the CLI interface and matching Python code instantly.

CLI Output Simulation

transcribe.py

1import gworker_client as gw2 3app = gw.App("whisper-batch")4 5@app.task(gpu="A100", timeout=600)6async def transcribe(audio_url: str):7    import whisper8    model = whisper.load_model("large-v3")9    return model.transcribe(audio_url)

< 0s

cold start

global regions

uptime SLA

idle cost

Compare & Estimate Costs

See how much you save using gworker's automatic scale-to-zero per-second billing compared to keeping a standard VM running 24/7.

Select GPU Instance

Active Compute Time60 hours/month

5 hrs (Minimal batches)360 hrs (12 hrs/day)730 hrs (24/7 continuous)

Traditional VMBilled 24/7 flatrate$840.96/mo

gworker ServerlessBilled per active second$69.12/mo

Estimated SavingsScale-to-zero efficiency$771.84/moSave 92%

See detailed per-second pricing across regions →

Loved by ML engineers

"Switched from wiring raw provider APIs to gworker in a weekend. We cut our GPU infra code by 80% and ship new models in hours, not days."

Alex R.

ML Lead, Acme AI

"The stage pipeline is the killer feature. We catch regressions in uat before anything touches prod. Transparent pricing is a bonus."

Sam T.

Senior Engineer, Lightspeed

"Per-second billing sounds like marketing until your idle bill drops to literally zero. This is how GPU infra should work."

Jordan K.

Founder, Inference Co.

Start building in minutes.

No credit card required. Free tier includes 100 GPU-seconds per day — enough to ship your first model.

Create free account View quickstart