Profile what you are about to train
Inspect full-dataset quality, preserve immutable versions, and keep content hashes attached.
Build tabular models you can trust.
Profile the full dataset, compare candidates, validate and explain the decision, then deploy on Kubernetes you control.
- Self-hosted
- Project-isolated
- Evidence-led
Project workspaceTraceable evidence
01DatasetProfiled and versioned
02CandidatesCompared with context
03EvidenceValidated and explained
04ReleasePromotion stays deliberate
05EndpointDeployed and monitored
Decision contextDataset, metrics, diagnostics, and lineage stay connected.
Ready for reviewPromotion stays in your hands
Sceptre turns a promising score into a reviewable operating decision.
Why evidence mattersOne path from raw table to governed endpoint.
Keep the data, experiment, model, and deployment record connected from the first profile to the final endpoint.
Choose with more than a score
Compare task-specific metrics, diagnostics, parameters, and experiment history in one reviewable record.
Make the decision reviewable
Validate on external data, inspect leakage controls, explain feature contributions, and retain audit evidence.
Operate the approved model
Register, promote, deploy, monitor, stop, and fall back from the same project workspace.
Replace handoffs with one accountable workflow.
Without Sceptre
Datasets, notebooks, and model files spread across tools
With Sceptre
One project record for data versions, experiments, models, deployments, and evidence
Every stage leaves the next team better prepared.
Evaluate candidates, control resource use, prove the decision, and deploy without changing tools or losing context.
Read the quick start01
Bring the data
Create a project, upload an immutable dataset version, and uncover quality risks before training.
02
Set the guardrails
Confirm the task and target, choose up to 20 candidates, and review the resource estimate.
03
Build the evidence
Compare task metrics, validate on external data, and explain the strongest candidates.
04
Operate the winner
Register, promote, deploy, monitor, stop, and keep a fallback ready from the same workspace.
Evidence follows the model
Monitoring reconnects outcomes to the original dataset.
Give every reviewer the same record.
Help technical reviewers, risk teams, and decision makers assess the same model from one traceable source of truth.
- Keep datasets, models, and predictions inside infrastructure you control.
- Package lineage, preprocessing, validation, and drift evidence for review.
- Promote, monitor, stop, or fall back through explicit actions.
Platform discipline without building the platform.
Give a small ML or data team a repeatable operating process without assigning engineers to assemble every MLOps component.
One accountable workflow
Keep profiling, training, review, and deployment connected.
Defensible model decisions
Put diagnostics, validation, explanations, and lineage beside the model choice.
Controlled shared capacity
Estimate demand before launch and cap every compute Job.
A durable operating record
Retain data versions, parameters, results, artifacts, and model status inside the project.
Know exactly where Sceptre fits.
Answers for teams evaluating a governed, self-hosted workflow.
What is Sceptre?
Sceptre is a self-hosted, Kubernetes-native tabular AutoML and MLOps platform. It connects dataset profiling, model training, comparison, validation, explainability, promotion, deployment, and monitoring evidence in one project-isolated workspace.
Is Sceptre self-hosted?
Yes. You install Sceptre on a Kubernetes cluster using its public Helm OCI chart. The bundled PostgreSQL, SeaweedFS, and MLflow services run inside your environment; cluster operators can also configure approved external services and exposure.
How is Sceptre different from other AutoML platforms?
Sceptre is designed for the work around and after model comparison. It keeps external validation, leakage controls, SHAP explainability, staged promotion, authenticated deployment, drift analysis, and downloadable audit evidence connected to the project.
Does Sceptre support model explainability?
Yes. Sceptre provides on-demand and cached SHAP explanations, historical model reconstruction, and support for non-predictive clustering estimators.
What data formats does Sceptre support?
Sceptre supports CSV, Parquet, Excel, JSON, and JSONL.
What machine learning tasks can Sceptre handle?
Sceptre handles classification, regression, clustering, and time-series problems with task-specific ranking metrics.
Who is Sceptre for?
Sceptre is built for small ML and data teams that already run Kubernetes, work primarily with tabular data, and need self-hosted model evidence and operations without assembling a general-purpose MLOps platform.
Does Sceptre train deep-learning or distributed models?
Not in the current release. Sceptre focuses on in-memory tabular estimators. Multi-node or distributed training requires additional infrastructure and a backend such as Dask or Ray.
Is Sceptre ready for production use?
The current release is documented for local development, evaluation, and compatibility testing. Production and shared-cluster environments require the controls in Sceptre's production-readiness guide.
Evaluate the complete workflow on your cluster.
Install the workspace, bring a tabular dataset, and follow one governed path from evidence to endpoint.