Compute API

Compute API for jobs that need to run reliably.

Submit workloads, set a max budget, and Badgr handles routing, capacity, health checks, logs, teardown, billing, outputs, and receipts.

Start with AI workloads. Run anything that needs controlled compute.

No setup fees. No monthly minimum. Pay only while your workload runs.

terminal

The control plane around every workload

Route

Badgr chooses the provider, GPU, or API route based on workload, budget, and availability.

Reserve

Your max budget is reserved before anything starts.

Start

Badgr launches the job or endpoint and checks it is healthy.

Observe

Logs, status, runtime, route, and cost are tracked while it runs.

Stop

Finished, failed, cancelled, or over-budget workloads are stopped automatically.

Prove

Outputs, logs, final cost, runtime, and receipts are saved.

Start with AI and GPU workloads

Five built-in job types to start. One API, one budget policy — Badgr handles the rest.

Image generation

Generate images via Stable Diffusion and similar models.

type: "image.generate"

Model serving

Serve open models as OpenAI-compatible endpoints with health checks.

type: "model.serve"

Custom GPU job

Run any script or container on H100, A100, L40S, or RTX 4090.

type: "custom.run"

LoRA fine-tuning

Fine-tune adapters on any base model with dataset and budget caps.

type: "train.lora"

Transcription

Whisper batch transcription with output and receipt on completion.

type: "audio.transcribe"

Start building

Read the docs →

What Badgr adds to every job

Submit work — Badgr handles the operational layer so you don’t have to.

Verified compatible capacity

Badgr confirms the GPU can run your workload before provisioning anything.

Max-cost protection

Set a spend cap before launch. The job stops automatically when it hits the limit.

Startup health checks

Badgr verifies the workload is running correctly before marking it ready.

Automatic teardown

Jobs stop when they finish or hit your limit, so billing never runs away.

Logs and failure receipts

Every run produces a full log and a receipt — even if the job fails.

Flexible routing across providers

Badgr routes to available capacity across providers so you don't chase inventory.

Two ways to submit jobs

CLI for quick runs, API for integrations. Both return the same job object.

CLI
$ badgr run python train.py \
--max-cost 10 --max-runtime 60
# or serve a model
$ badgr serve meta-llama/Llama-3.1-8B-Instruct \
--max-cost 20
API (curl)
curl https://aibadgr.com/v1/jobs \
-H "Authorization: Bearer $KEY" \
-d '{
"type": "custom.run",
"input": {"script": "python train.py"},
"policy": {"max_cost": 10}
}'

AI API

Need one endpoint for model calls?

Route OpenAI, Claude, Gemini, and open-model requests through a single OpenAI-compatible endpoint with retries, fallback routing, spend visibility, and request-level receipts.

See AI API →

Fallback routing

Fail over to a backup model automatically.

Request-level receipts

Every call gets a cost, route, and latency record.

OpenAI-compatible

One base URL swap. Your SDK keeps working.

Available GPU capacity

Browse Badgr capacity updated every 15 minutes. Set max-cost protection before launch.

Badgr H100

$1.90/hr

80GB

vLLM, long context, production

Available now

Badgr A100

$1.40/hr

80GB

Training, inference, batch

Available now

Badgr 4090

$0.45/hr

24GB

ComfyUI, LoRA, batch jobs

Available now
See all available capacity →

Need dedicated or larger GPU capacity?

Request A100, H100, or L40S capacity for multi-GPU workloads or longer runs. No login required. Quote within one business day.

Request a quote →