One click
to one million
RL environments.

Isaac Sim training on cloud GPUs. Start training Read the docs
We Support
Isaac Sim
Weights & Biases Live logging
TensorBoard Logs

Five lines to train.

pip install, submit, wait, download.

from canard import Client # Connect to Canard client = Client(api_url="https://api.canard.cloud") # Submit a training run run = client.submit_training_run( name="Go2 Standing", num_envs=8192, max_iterations=5000, gpu_count=2, ) # Wait and download results run.wait_for_completion() run.download_results("./results")
Why Canard

Your GPU vs. ours.

Canard
Local GPU

Scales to thousands of envs

Spin up 8,192 parallel environments in seconds. No hardware limits.

Parallel parameter sweeps

Run 50 configs at once. One function call, all results in parallel.

Live dashboard & logging

Reward curves, GPU utilization, W&B and TensorBoard built in.

Auto checkpoints to cloud

Every checkpoint saved. Resume or download the best model in one call.

Reinstall NVIDIA drivers 4 times

Debug CUDA version mismatches at 2am. A rite of passage.

Pay only for what you use

Scales to zero when idle. No hardware depreciation, no electricity bills.

Pricing

Pay for GPU time. Nothing else.

Standard

General purpose
$0.50 /hr
Best value. Great for prototyping and most training workloads.

Fast

High performance
$1.50 /hr
For large observation spaces, big batch sizes, and deadline-driven sweeps.
Coming Soon

Turbo

Maximum throughput
TBD /hr
Multi-GPU clusters for the largest training runs. Stay tuned.

Enterprise

Volume discounts on GPU compute.

Contact us

Stop managing GPUs.
Start training policies.

From $0.50/hr. No subscriptions. No platform fees.

Start training