Valohai
Turku platform that versions every ML run, pipeline and dataset, and runs on any cloud, on-premises or air-gapped
Quick Overview
| Company | Valohai Oy |
|---|---|
| Category | MLOps Platforms |
| Headquarters | Turku, Finland |
| Founded | Not stated |
| EU Presence | Yes - Finland (EU) |
| Data Location | Managed cloud, your own cloud accounts, on-premises or air-gapped |
| Open Source | No |
| GDPR Compliant | Yes |
| Self-Hosting | Yes - on-premises, private cloud and air-gapped deployment |
| Main Features | Experiment tracking, hyperparameter optimization, data versioning, pipelines with quality gates, triggers, Model Hub, Git-based workflows, cost tracking, SSO access control |
| Pricing | Per-user licence, quote on request; unlimited projects, experiments, pipelines and deployments |
| Best For | ML teams that need reproducible runs on their own cloud or GPU cluster |
| Replaces | Weights & Biases, SageMaker, Vertex AI |
Detailed Review
Valohai wraps each training job so that its code, data, parameters and environment are recorded together. You keep ordinary Python scripts that read from and write to local folders, and the platform syncs them with cloud storage and versions every run.
The terms are issued by Valohai Oy, Linnankatu 16, Turku, and the privacy policy names the Finnish Data Protection Ombudsman as the supervisory authority. The home page says the company has operated since 2016.
Reproducibility as the core feature
Every run is versioned automatically and traceable back through the pipeline to the exact dataset and preprocessing step, and forward to the deployments that use a model version. The site describes one-click reproduction of an earlier run.
Pipelines cache each step separately, so only changed steps rerun, and conditional logic or manual approval gates can block a model that does not beat its baseline.
Your infrastructure, not theirs
Jobs can run on virtual machines, Kubernetes or Slurm, on any cloud or in your own data centre. The pricing page lists managed cloud, on-premises, private cloud and air-gapped deployment, plus Dynamic GPU Allocation for on-premises hardware.
Partners listed include AWS, Microsoft Azure, NVIDIA, Oracle, OVHcloud and Snowflake.
Per-user pricing
Valohai charges a fixed fee per user per month regardless of how many projects, experiments, pipelines or deployments you run. The price per user is not published, so you request a quote.
That model suits busy teams with few users and punishes the opposite pattern of many occasional contributors.
Limitations to Consider
The per-user price is not published, so you cannot compare it with Weights & Biases without a sales call. The platform expects a specific code layout (files under /valohai/inputs and /valohai/outputs, parameters through argparse, metrics printed as JSON, a valohai.yaml), which means a small migration and some platform coupling. Its public terms are version 8 from 2021. It is an execution platform, so team-facing dashboards are lighter than a dedicated tracking tool.
Who Should Use Valohai
European ML teams that train on their own GPUs or several clouds and want every run reproducible, from a Finnish vendor.
Pros and Cons
Pros
- Every execution versioned with full lineage
- Runs on any cloud, on-premises or air-gapped
- Pipelines cache steps and support approval gates
- Finnish company with published Turku address
- Unlimited projects and runs under one per-user fee
Cons
- Price per user not published
- Code must follow Valohai file and metric conventions
- Terms page dates from 2021
- Lighter dashboards than dedicated tracking tools
Alternatives to Valohai
Other European mlops platforms worth comparing before you decide:
ZenML
Germany · Munich open source pipeline orchestration with artifact and model tracking
Hopsworks
Sweden · Stockholm feature store with model registry and serving, free tier
DagsHub
Israel · Experiment tracking, data versioning and annotation on one platform
UbiOps
Netherlands · Hague platform to deploy and orchestrate models on any infrastructure
See all European mlops platforms or compare them directly against Weights & Biases on our alternatives page.
Frequently Asked Questions
No. The pricing page describes a per-user licence with unlimited projects, experiments, pipelines and deployments, and sends you to a custom quote for the amount.
Yes. The pricing page lists managed cloud, on-premises, private cloud and air-gapped deployment, and jobs can run on virtual machines, Kubernetes or Slurm.
The site describes the change as small: rewrite file paths, lift parameters into argparse, print metrics as JSON and add a valohai.yaml.