Sign in Start free

Pricing

Start free. Grow when your models do.

The free plan is enough to run the entire Titanic walkthrough. Team and Enterprise add the limits, deployment options and support that production workloads need.

Free

For evaluation
$0/ month

Everything you need to upload a model, explain it and watch it in production.

Start for free
  • 10 models in the registry
  • 500 explanations a month
  • 20 analyses a month
  • 5,000 logged predictions a month
  • Python client, REST API and the dashboard
  • Remote inference and private weights
  • Community support on GitHub

Team

Most teams
Custom

Higher limits and shared projects for teams putting several models into production.

Talk to us
  • Everything in Free
  • Higher monthly limits on models, explanations, analyses and predictions
  • Shared projects across your organisation
  • Higher rate limits for batch explanations
  • Email support

Enterprise

Regulated workloads
Custom

For organisations that need Kaimei inside their own perimeter, with procurement and security review.

Contact sales
  • Everything in Team
  • Self-hosted or private deployment of the Kaimei API
  • Volume pricing and custom limits
  • Security questionnaires and procurement support
  • A named contact who knows your models

What counts

Four things are metered. Here is what each one means.

Model

One entry in the registry, with all of its versions. Re-uploading a model replaces it, along with what it logged, so re-running a notebook does not add to the count.

Explanation

One call that explains one input, by any method, including counterfactuals and image heatmaps. Batches count per input.

Analysis

One study of a whole model against its reference data: importance, rules, behaviour curves, interactions, robustness and drift.

Logged prediction

One prediction your stack made, or Kaimei made for you, stored with its inputs and output so it can be explained and monitored.

Hourly rate limits also apply, for example 120 explanations an hour on the free plan. 429 responses carry a Retry-After header and the Python client honours it automatically.

Questions

Frequently asked

Does the free plan really cover the Titanic walkthrough?
Yes. Running all five notebooks creates four models (five at most while an endpoint runs), 27 explanations, five or six analyses and about 1,560 logged predictions. Re-running them replaces what they created rather than adding to it.
Does Kaimei store my model file?
No. The client reads the fitted estimator in your process and uploads its layers and weights as a structure. Kaimei never receives the original pickle, .h5 or .pt file, and it does not host a binary artifact. With weights="summary" only shapes and statistics are uploaded.
My model is not in the supported list. Can I still use Kaimei?
Yes, in two ways. Register an HTTP scoring endpoint and Kaimei explains and monitors the model through it. Or compute attributions yourself, with TreeSHAP for example, and POST them. The fourth Titanic notebook does both with XGBoost.
Can I run Kaimei inside my own infrastructure?
Private and self-hosted deployments are part of the Enterprise plan. The Python client already supports a custom base URL, so code written against the hosted API moves without changes. Talk to us about your requirements.
How do I get support?
Free plan users open an issue on GitHub. Team and Enterprise plans include email support and, for Enterprise, a named contact.

Not sure which plan fits?

Start on Free today. When you outgrow it, tell us about your models and we will size a plan around them.

Start for free Talk to us