Valohai

Valohai is a machine learning operations (MLOps) platform-as-a-service that automates the infrastructure around training, evaluating, deploying and managing machine learning models. Core product areas include pipelines that cover both experimentation and production with per-step caching, conditional logic and quality gates, parallel task execution (from four to over 100 parallel executions), and human-in-the-loop approval steps; versioned, immutable datasets that are cached across executions and versioned incrementally by added and removed files, with aliases used to promote a version to production; automatic lineage and traceability linking models to the executions, code versions, parameters, dataset versions and deployments involved; and an operations dashboard that combines compute spend, quality metrics and infrastructure utilization in one view.

The platform also targets LLM and RAG workloads, offering systematic sweeps across chunk sizes, embedding models, retrieval strategies and prompts, multi-model comparison reporting quality, latency and cost per token, a `valohai-llm` Python package that can post results from any environment, and tracing integration with Langfuse. Additional capabilities listed by the company include experiment tracking, hyperparameter optimization, data versioning and tagging, CI/CD pipelines and triggers for ML, a Model Hub for model versioning, distributed training, cost and usage tracking with underutilization alerts, access control integrated with SSO, encryption at rest and in transit, and an audit log for governance. Valohai positions itself as framework- and language-agnostic (Python, R, TensorFlow, PyTorch, scikit-learn and others) and orchestrates workloads on virtual machines, Kubernetes, Slurm or other available technologies.

Valohai emphasizes low code intrusion: user code reads and writes files in a local directory, takes parameters via argparse and prints metrics as JSON, while the platform syncs files with cloud storage and versions each run. The company publishes "Agent Skills" for coding assistants such as Claude Code, Cursor, Copilot, Codex, Gemini and Zencoder to automate migration of existing ML projects to this convention.

Founding story

Valohai was founded in 2016 by a team of engineers whose backgrounds included various software development companies, among them San Francisco-based Leap Motion. The founding team comprises CEO Eero Laaksonen, CTO Aarni Koskela and Ruksi Korpisara (machine learning). The founding premise was that companies beginning large-scale machine learning had to build extensive overhead infrastructure — a setup period the company said could take as long as two years — before applying deep learning to their actual problem, and that a standardized infrastructure and workflow would let teams focus on models instead.

Business model

Valohai sells subscriptions to its MLOps platform on a per-user license model: a fixed fee per user per month regardless of usage volume, with unlimited projects, experiments, pipelines and deployments included. Subscriptions bundle advanced security and access control, technical support from a customer engineering team, and training and onboarding resources including the Valohai Training academy. Prices are not published; customers obtain custom quotes through the sales team.

Recurring per-user subscription licensing, with pricing decoupled from the amount of compute or number of workloads run on the platform; quotes are tailored by the sales team rather than listed publicly.

Traction

Valohai reports customers across Europe, Asia and the USA and states it has helped data science teams build, test and deploy deep learning models 10x faster. Customer voices published by the company include co-founder/CTO testimonials citing avoided months of in-house infrastructure work and hybrid-cloud portability, Reaktor's use of the platform on an autonomous model ferry project, and use by AI consultancy Silo.AI. The company team page lists roughly 20 named employees plus four advisors.

Latest developments

Recent company updates include the launch of a productivity/operations dashboard giving combined visibility into ML operations (August 2025) and availability of the MLOps platform on the Oracle Cloud Marketplace (September 2025). Product messaging has shifted toward supporting AI products that pair LLMs with specialized models, including RAG and LLM evaluation sweeps, multi-model cost/quality comparison, a pip-installable `valohai-llm` package, Langfuse tracing, and Valohai Agent Skills for coding assistants such as Claude Code, Cursor and Copilot to migrate existing ML projects onto the platform. Earlier 2025 content covered managing massive datasets and integrations with NVIDIA MONAI and NVIDIA NeMo.