Continuous Integration for Machine Learning

Turn your GPUs into build servers.

(No credit card information required.) See pricing information.
  • VCS Integration

    Losswise integrates with Github, Bitbucket, and more. Every time you commit code, Losswise triggers a build.

  • Provisioning

    Losswise agents run builds on your servers, whether they are on-prem or in the cloud. Super easy agent setup.

  • Monitoring & Analytics

    Visualize logs, loss functions, and predictions. Losswise provides super simple API's to track ML performance.

  • ~ (zsh)> git commit -m "Smarter CNN's [lw]"> git push origin train/lenet-1


    It starts with a commit.

  • Build

    Build gets dispatched to agents.

  • Collect

    Collect metrics in realtime.


Results are compiled in searchable, sortable, and explorable tables, so you can figure out what works and what doesn't.

    Built by deep learning experts

    We built Losswise to make it easy to track the progress of a machine learning project. With Losswise, we find we spend less time worrying about the operational complexity of training models so we can focus on other things such as improving datasets or experimenting with new model architectures.

Learn more at

Get started today.

Choose the plan that's right for you and your team.

  • Pro Plan

    • Unlimited experiments
    • Unlimited users
    • 1 month free trial (no card required)
    • First 20 hours per month are free
    • $0.2 / hour of train time
    • 3 month data retention
  • Premium Plan

    • Unlimited experiments
    • Unlimited users
    • 1 month free trial (no card required)
    • Up to 2000 hours of training time
    • $0.1 / hour of training time overage
    • Unlimited data retention
  • Enterprise plan

    • Custom on-prem dashboard
    • Pay by invoice
    • Email and phone support


What is considered an hour of training time?

One hour of train time is the total amount of time tracked by Losswise. If you run 2 experiments at the same time for 30 minutes, that counts as 1 hour of training time. We use total training time per month in order to represent the total number of experiment time logged by Losswise. We charge per hour of experiment time because this way users only pay for what they use.

What are some examples of Pro Plan calculations?

If you have 1 GPU, and you have it running with Losswise at 50% utilization for 1 month, that counts as 0.5 * 730 hours in one month = 365 hours of training time. Since the first 20 hours per month are free, the total fee ends up being (365 - 20) * 0.2 = $65. If you run 4 GPU's at 50% utilization for 1 month, the cost ends up being: (0.5 * 730 * 4 - 20) * 0.2 = $288.

What are some examples of Premium Plan calculations?

If you use less than 2000 hours of training time in a given month, the fee is exactly $200 per month. If you use 3000 hours of training time your cost is $200 + 1000 * 0.01 = $300. And so on.

What is Losswise's security and privacy policy?

Losswise is obsessed with keeping customer's data secure and private. Our servers are hosted on AWS and we follow AWS best security guidelines. We do not access anyone's metrics or logs unless it's required to solve a customer problem. The architecture of our product allows users to keep code and data on their servers and lets us handle the metrics and logs, following the principle of least privilege.

Do you have plans for research universities?

Yes, contact us directly and we'll create a custom plan that works for your needs.

Can I cancel anytime?

Absolutely! You can always cancel and we will send you an email with all your metrics data if you wish.