Build artificial intelligence that anybody can use.

Catacomb is the simplest machine learning platform for deploying prototypes, conducting quality assurance, and benchmarking models.

Compatible with any platform or framework:

Accuracy on a single dataset isn't enough.

The only way to know if your models are safe and useful is to have real people use them, from end-users to domain experts.


Simple Deployment

Catacomb automatically creates Dockerfiles, front-end code, and even REST APIs, so that you don't have to.

View Catapult

Built-In Collaboration

Work as a team by comparing models, sharing examples, and providing feedback with no code required.

View Catalogue

Easy Integration

Catacomb only requires a few lines of Python to support your current workflow and model architectures.

Read Quickstart Guide
svg svg

Understand your models before releasing them to the world.

Example Notebook Demo


up to 10 GB

Intuitive infrastructure.

The most advanced cloud platform for machine learning testing and quality assurance, made to integrate seamlessly with any stack.

  • Model Version Controlling
  • Continuous Integration Support
  • Collaboration Suite
  • Custom Metrics and Evaluation Datasets
  • System Comparison Tools

Pitch-Ready Demos

Instantly create an interface for your clients to query your models, browse historical performance, and compare against the status quo.

Fine-Grained Testing

Identify inflammatory or problematic behaviour in models by easily specifying custom input sets and metrics for evaluation.

Shadow Deployments

Stream inputs that your production models are receiving to your prototype models, and compare their results.

Automated Data Synthesis

Generate realistic data from your test inputs to understand model performance in multiple different scenarios.

Pick a plan and iterate faster.

Affordable plans for any organization, from research labs to industry AI teams.


Get started now.

Get Started


Flexible pricing at any scale.

View Pricing

Any questions?

Artificial intelligence is hard. We'll be here with you every step of the way.

Leaving your machine learning models in a Python notebook doesn't do them justice. Catacomb lets you see how your models respond to real-world inputs that might not exist in your training sets, so you can interate faster without compromising the reliability of your production systems.
Yes! Catacomb is meant to unobtrusively be another layer of your data science stack, and is compatible with most machine learning platforms.
Not yet. Currently, Catacomb provides low-fidelity hosting for testing and quality assurance that is meant to work in tandem with other production-specific services.
The free tier includes unlimited public hosting, as well as other evaluation features. Refer to our pricing page for more information on our flexible pricing.
Still have a question? Reach out to us here