Tracking and registry
Keep experiments, runs, models and approval stages available.
WZ-IT operates MLflow as a traceable ML platform with a tracking server, database, artifact storage, model registry and secured access.
Companies worldwide trust WZ-IT
The following are trademarks of their respective owners: MLflow (the MLflow project and The Linux Foundation). WZ-IT is an independent service provider and has no business, partnership, or contractual relationship with these companies. We offer independent migration, installation, hosting, and operations services.
MLflow records parameters, metrics, artifacts and model versions. For teams, tracking, storage, access, retention and deployment paths must be designed together.
We assess frameworks, data volumes, CI/CD, target environments and compliance requirements and turn them into an operable architecture.
Training data, compute, feature stores, inference and business model approvals remain separate architecture components. We define responsibilities and data paths clearly.
The calculator covers standard operations. Artifact storage, CI/CD, training environments and special integrations are sized separately in the proposal.
Tracking, backend store, artefact storage, registry and connected training systems require consistent security and recovery.
Keep experiments, runs, models and approval stages available.
Protect database and object storage consistently.
Connect CI/CD, notebooks, pipelines and serving systems.
Control authentication, retention, monitoring and updates.
WZ-IT operates MLflow and its stores; model quality and business approval remain with customer teams.
| Area | Responsibility | Scope and boundaries |
|---|---|---|
| MLflow services | WZ-IT | Deployment, proxy, backend connection, monitoring and updates. |
| Backend and artefact store | WZ-IT | Configuration, encryption, backup and restore. |
| CI/CD and training | Shared | We integrate; the customer supplies pipelines and runtime context. |
| Model stages | Shared | Technical roles and gates implement the approved process. |
| Datasets and quality | Customer | Training, evaluation and risk remain customer-owned. |
| Automation | Optional WZ-IT service | Deployment and evaluation pipelines are scoped separately. |
Compare parameters, metrics, runtimes and results across training and evaluation runs.
Store model files, charts and other results with versions in a connected artifact store.
Move registered models through controlled testing, approval and deployment processes.
Make runs and model versions centrally visible and document responsibilities transparently.
Connect training, CI/CD, object storage and target platforms through supported interfaces.
Protect the tracking server, artifacts, database and administrative paths from unauthorised access.
Assess existing data and configuration, migrate them in a test run and move to managed operations through a controlled cutover.
Secure SSO, roles, administrative paths and external access for the application and existing infrastructure.
Back up all stateful components consistently and document the recovery path for the agreed scope.
Monitor and update the application and its technical dependencies and operate them under the agreed service level.
A clearly defined operating scope instead of an opaque hosting flat fee.
We set up a test instance for you, usually on the next business day. No payment details required. After seven days it is deleted unless you continue.
We combine the right compute size with ongoing operations, backups, monitoring and a service level appropriate for the criticality of MLflow. High availability and recovery targets are designed separately where needed.
We also design custom hosting architectures, integrations and migrations around MLflow. Contact us for a technical assessment.
One managed standard MLflow application is included in the Starter workload. Business and higher levels add a flexible operations allowance for planned work during regular service hours. Select compute, additional applications, storage and the appropriate service level.
A workload is one compute instance with the applications agreed for it.
One standard app per workload is already included. Additional dedicated servers count as separate workloads.
€79.90 per started TB and month, including daily encrypted offsite backup with 7-day retention.
Enquiry
Briefly describe the current state and objective for MLflow. We assess infrastructure, integration, and ongoing operations.
Experiment count, artefact volume, retention and automation determine platform capacity.
| Usage scenario | Technical starting point | Key factors |
|---|---|---|
| Small data-science team | Database and object-store assessment | Tracking, registry and artefacts are sized separately. |
| Many runs or large artefacts | Storage and retention design | Lifecycle and backup windows are decisive. |
| Automated training | API and load assessment | Concurrent runs and CI/CD are considered. |
| Production model approval | Governance and recovery design | Roles, audit and recovery are planned. |
Runs, artefact volume, retention, identity and integrations are reviewed before a proposal.
MLflow should be close to training data, artefacts and pipelines.
Tracking and registry with European object storage.
Integration with existing training and storage services.
Operate near local GPU, notebook and data systems.
Central metadata with controlled artefact and training routes.
MLflow references external artefacts and runtimes; every component needs a recovery path.
Data scientists, CI/CD, notebooks and automated jobs.
TLS, identity, APIs and restricted service credentials.
Access to experiments, models and production stages.
Tracking, registry, metadata and lifecycle interfaces.
Runs, metrics, tags and registry state.
Models, charts and versioned outputs.
External compute and deployment systems.
MLflow does not replace business model validation or organisational release controls.
Answers about tracking, artefacts, registry and operations.
The agreed tracking and registry stack including database, artefact store, proxy, monitoring and backup.
No. Training and GPU resources are connected or quoted separately.
Artefact store and backend database are backed up with coordinated recovery points.
Yes, especially when training data and GPUs are already local.
Yes. Credentials, stages, tests and deployment targets are scoped explicitly.
As an alternative to managed hosting in the data centre, WZ-IT provides the hardware, configures MLflow, and handles hardening, monitoring, updates, backup and technical support. Access can be limited to the internal network or enabled through VPN and existing identities.
from EUR 349 excl. VAT / month · plus one-time provisioning and initial setup

01.06.2026
Your own ChatGPT, running entirely in-house: a familiar chat interface, a large language model on your own hardware behind it, without a single prompt going...
24.05.2026
Anyone bringing AI into production business processes quickly faces an uncomfortable question: what is actually happening in there? Which prompt went to which model, why...
05.05.2026
The EU AI Act's high-risk obligations do not start on 2 August 2026. Stand-alone high-risk systems under Annex III now have until 2 December 2027,...
These solutions are often used together with MLflow
These solutions offer similar functionalities and can be evaluated together
These solutions are direct alternatives with similar use cases
No risk: worst case, you leave with a clearer understanding of your project than before.


“WZ-IT's advice on our Azure migration was technically sound and completely non-binding right from the intro call - we took away a great deal.”
Whether a specific IT challenge or just an idea - we look forward to the exchange. In a brief conversation, we'll evaluate together if and how your project fits with WZ-IT.