Changelog
What changed, release by release, written for the people who use the platform rather than the people who build it.
- GeneFlow
Model serving with drift monitoring
- Deploy a registered model as an endpoint with rolling updates, autoscaling, and readiness checks.
- Sampled inference logging and PSI-based drift detection against a saved baseline, with an acknowledge workflow for alerts.
- Endpoint dashboard: live QPS, p95 latency, error rate, and cost per endpoint, with a one-click drift acknowledgement banner.
- Python: geneflow.serving — create and update endpoints, log inferences, save baselines, check and acknowledge drift.
- GeneFlow
Import from MLflow, and observability for the runner
- One-click MLflow importer: bring experiments, runs, artifacts, and the model registry across from an existing MLflow server.
- Metrics for the collaboration gateway and job runner, with a ready-made dashboard.
- End-to-end soak test across tracking, artifacts, registry, prompts, evaluation, lineage, and projects.
- GeneFlow
Kubernetes job runner, real-time collaboration, CLI
- Projects can run locally, in Docker, or as Kubernetes jobs with the same definition.
- Real-time collaborative editing for prompts, backed by CRDTs so two people can edit without conflicts.
- Command-line tool for auth, runs, models, and deployments.
- GeneFlow
Version compare and the MLOps control plane
- Side-by-side model version comparison with cost and metric deltas.
- Four-tab control plane for experiments, registry, prompts, and evaluation.
- Artifact upload and download to S3 or local storage; MLflow Projects compatibility.
- Agent tools for ML, MLOps, AI-engineering, and data-science roles.
- GeneFlow
GeneFlow launch: tracking, registry, prompts, and the Python SDK
- Experiment tracking that is MLflow-compatible at the wire level — existing mlflow.start_run() code works unchanged.
- Model registry with approval gates; a first-class prompt registry with vector search; lineage across runs, datasets, and features.
- REST API with both MLflow-compatible and GeneFlow-native endpoints; Python SDK.
- Platform
Spring release: role surfaces, data-subject requests, PDF reports, observability
- Fifteen production role surfaces, so each job function lands on a home page built for its work.
- Data-subject access and deletion requests handled by a purge worker with an audit record.
- Report Builder renders real PDFs; batch jobs dispatch to Kubernetes.
- Alerting and observability infrastructure, with per-route integration tests against a real database.
- Workspace isolation is enforced structurally in the data layer rather than by convention.
Try it
Everything above is in the product today.
Request access, or read the documentation that ships with each release.