getML
getML is a machine learning suite that automates feature engineering on relational and time-series enterprise data. Five supervised feature-learning algorithms build predictive pipelines from raw multi-table data, driven entirely through a Python API installed on your own infrastructure.
What is getML?
getML is a machine learning suite for predictive analytics on relational and time-series data, published by Code17 GmbH in Leipzig. Its purpose is narrow and clearly stated: to remove the manual feature engineering step that stands between a corporate database and a working predictive model. At its heart are five feature-learning algorithms that automate that step using supervised learning rather than brute-force templates. Technically, the approach generalises gradient boosting to multi-relational decision trees, which lets supervised learning operate directly on raw relational data instead of a pre-flattened table. An iterative, tree-based search runs inside the relational structure to find the best combination of aggregation functions and conditions for a given target variable. The suite has three components: a Python API, an Engine, and a Monitor shipped alongside the Engine. The programming model borrows its shape from scikit-learn. A DataContainer holds the population and peripheral tables together with the train, test and validation split; a StarSchema or DataModel object captures the relationships between them; feature learners and predictors are declared separately and bundled into a Pipeline whose fit, score and predict methods drive the whole process. The homepage claims three lines of Python are enough to learn from billions of candidate features. Two editions share that same API. The Community edition is open source under the Elastic License v2 and ships the FastProp learner; the Enterprise edition adds Multirel, Relboost, RelMT and Fastboost, along with memory-mapped ScaleGBM predictors for datasets that do not fit in RAM. Published figures on the homepage claim a tenfold speed-up in customer projects and accuracy gains of up to 65 percent over baseline models. The company was founded in 2017 and describes a team of more than ten engineers in Leipzig and Munich. The most recent major release, getML 1.5, was announced in November 2024 after nine months of development, and the Python package stands at version 1.5.1.
What it does
- Generate features automatically from raw relational data, without writing feature engineering code
- Learn across several linked tables at once, without manually flattening them into a single view
- Train a predictor and assemble an end-to-end prediction pipeline in a few lines of Python
- Transpile the learned features into readable SQL so they can run in production without the engine
- Search hyperparameters with random search, Latin hypercube or Gaussian optimisation
- Track experiments and artefacts through the MLflow integration
- Handle datasets larger than available memory using memory-mapped predictors
When to use getML / When not to
A quick filter to help you decide if getML is the right fit.
When to use getML
- Data science teams modelling multi-table relational schemas, where hand-written feature logic normally takes months
- Analysts and engineers forecasting on enterprise time series who have outgrown Prophet, tsfresh or FeatureTools
- Organisations whose data cannot leave their own infrastructure, since the engine runs entirely on the customer's machines
- Python-fluent practitioners comfortable with a code-first workflow and no graphical assistant
- MLOps and platform engineers who need trained features transpiled into readable SQL for production deployment
When not to use getML
- Teams looking for a hosted SaaS product: nothing runs on the vendor's servers, you install and operate the engine yourself
- Non-technical users expecting a no-code interface, since every operation goes through the Python API
- Anyone working on free text, images or audio: the suite addresses tabular relational and time-series data only
- Buyers looking for generative AI or large language models, which the product does not touch
- Organisations that need a published price before starting a conversation, as the Enterprise edition is quoted only through sales
How to use getML
A typical end-to-end flow, from setup to results.
- Check the prerequisites: a 64-bit system, Python 3.8 to 3.12, and glibc 2.28 or later for a native Linux install
- On Linux, run pip install getml, which installs both the Python API and the Engine natively
- On macOS or Windows, set up the Docker container that runs the Engine and install the Python API with pip
- Alternatively, build the API or the Engine from source, or install from a versioned archive
- Follow the Quick Start, announced as a five-minute introduction to feature learning on a simple example
- Work through the two-hour Walkthrough, then the Concepts section for the underlying theory
- Stage your data in a DataContainer, including the train, test and validation split
- Declare the relationships between your tables with a StarSchema, or a DataModel for arbitrary complexity
- Choose feature learners and a predictor, bundle them into a Pipeline, then call fit and predict
- Move to production by transpiling the features to SQL, using the Enterprise HTTP endpoints, or wrapping the pipeline in a FastAPI service
Pros & Cons
Pros
- Removes the writing and maintenance of feature engineering code, which is the stated bottleneck the product targets
- The vendor claims a tenfold speed-up and up to 65 percent accuracy gain over baseline models on customer projects
- The Community edition is genuinely open source under the Elastic License v2, with demo notebooks runnable locally or on Google Colab
- Both editions share the same Python API, so a prototype built on Community moves to Enterprise without rewriting
- Everything runs on the customer's own infrastructure, so training data never leaves it
- Learned features are transpiled into readable SQL, making them auditable and deployable without the engine
- Broad coverage of databases and file formats on the input side, and documentation split into three depths of detail
Cons
- No price is published anywhere: the site has no pricing page at all, and the Enterprise edition is quoted only through sales
- There is no graphical interface outside the Enterprise web frontend; everything is done in Python
- The Community edition is limited to Linux and Docker and has no hyperparameter optimisation
- The Enterprise trial is advertised but its download page still describes the 1.5.0 build as being in the final stages of testing
- The trial terms page on the site is published with a title and no body at all, and the real terms sit on a third-party Notion domain
- The privacy statement's own body says it covers a different website, and carries no resolved update date
- Freshness signals are weak: the last blog post dates from December 2024 and documentation pages carry September and October 2024 dates
Pricing & Plans
A permanent free plan exists. The getML Community edition is open source under the Elastic License v2, carries no time limit and is installed with a single pip command; community support is provided free of charge through the GitHub issue tracker and discussions, with no guaranteed response time. The Enterprise edition is proprietary and its price is not published: no figure appears anywhere on the site, the sitemap of 157 pages contains no pricing page, and the only route to a quotation is the Talk to sales form. Enterprise support is likewise described as negotiable, with guaranteed response times and a negotiable volume. A free, non-commercial trial licence for the Enterprise edition is also offered.
- free
- Elastic License v2
- Linux and Docker
- FastProp feature learner
- base preprocessors
- linear
- logistic and XGBoost predictors
- transpilation to readable SQL
- proprietary licence
- price not published
- includes everything in Community plus Multirel
- Relboost
- RelMT and Fastboost
- the Mapping preprocessor
- ScaleGBM predictors
- transpilation to seven SQL dialects
- free
- no guaranteed response time
- GitHub issue tracker and discussions
- negotiable
- 100 percent product coverage
- guaranteed response times
- private issue tracker and datasets
- negotiable volume
- flexible scope covering deployment and MLOps advice
- hands-on data science support and API training
Data, GDPR & hosting
A consolidated view of how getML handles your data.
GDPR overview
GDPR implementation is documented, but only for the website. The privacy statement names Code17 GmbH, Philip-Reis-Strasse 11b, 04179 Leipzig, with a phone number and contact@getml.com as controller, and states that personal data is processed in compliance with the GDPR. It cites Article 6(1)(a) to (f) as legal bases, lists the rights of access, rectification, erasure, restriction, portability, objection and withdrawal of consent, and the Article 77 right to complain to a supervisory authority. Fifteen recipients are named, among them Cloudflare, Google Ireland, HubSpot, Notion and Framer. Transfers to the United States and other third countries are acknowledged under Articles 46 and 49. No data protection officer and no Article 27 representative are designated, and no data processing agreement is offered. The document's own body says it covers the website code17.io.
Who owns the data?
The published documents say almost nothing about data ownership, and that silence has a structural explanation: getML is installed and executed on the customer's own machines, through pip on Linux or a Docker container elsewhere, so the vendor never receives the data being modelled. The trial licence agreement governs only the right to use the software; it contains no clause on data, output or intellectual property in the customer's datasets. The privacy statement published by Code17 GmbH covers the website alone and names the company as controller for visitor and applicant data. No data processing agreement is published or offered anywhere on the site.
Reuse rights
Nothing in the published terms restricts what customers may do with their own data or with the models and features getML produces from it, because the trial licence agreement addresses only the software itself. It grants a non-transferable right to use the product for evaluation, testing and inspection, forbids passing the software to third parties and forbids commercial or production use during the trial; it says nothing about outputs. Features learned by the engine can be transpiled into readable SQL, which means the result of a training run can be lifted out and run without the product. Personal data collected through the website is used for contact handling, contract preparation, job applications, advertising, statistical analysis, website security and social-media communication, under consent, contract, legal obligation or legitimate interest. No automated decision-making or profiling with legal effect is carried out.
Data retention & training
Hosting summary
The product itself hosts nothing. getML is installed and executed on the customer's own infrastructure, natively on Linux or inside a Docker container elsewhere, so the vendor receives no customer data and declares no hosting country or region for it. A targeted search across the whole collected corpus found no mention of a data centre, a hosting location or a jurisdiction for product data. For the public website, the documentation is served from GitHub Pages behind Cloudflare DNS, while the privacy statement names Framer as the website host, which matches the publisher's corporate site rather than the documentation portal. That same statement acknowledges transfers of website and applicant data to third countries, including the United States, relying on Articles 46 and 49 of the GDPR, and notes that some destinations have no adequacy decision. Fifteen recipients are named, several of them United States providers.
Things to keep in mind
Risks and trade-offs to weigh before adopting getML.
- The domain is second-hand: getml.com was registered in July 2011 and its earliest web archive capture, from March 2000, serves an unrelated Japanese site. Three separate timelines coexist - domain, company and product - and none should be read as the others
- The legal entity was registered on 1 September 2017 under a different name, The SQLNet Company GmbH, and only became Code17 GmbH in October 2023. Any search on the current name against a pre-2023 source can fail for the wrong reason
- The privacy statement's own body says it applies to the website code17.io, although it is published and linked as the privacy policy of getml.com. Any contractual fact drawn from it is therefore revocable
- The two Notion pages linked as Privacy Notice and Website Data Privacy Statement carry the same text, and neither shows a resolved last-update date, so the currency of the published terms cannot be checked
- The trial terms page on the site itself is published with a heading and no body, and the Enterprise trial download is still announced as pending - do not assume a self-service trial is available today
- No certification, no data processing agreement and no product security commitment is published, and the site never states whether customer data is used to train models. Absence of a claim is not a guarantee
- This is code-first software you must install, operate and keep running yourself; treating it as a managed service will underestimate the engineering effort, and its automated features still require the judgement of someone who understands the underlying data
Setup & Integrations
Technical difficulty
Moderate, and squarely for technical users. On Linux a single pip install brings in both the Python API and the Engine; on macOS or Windows you must first set up a Docker container for the Engine. Prerequisites are a 64-bit system, Python 3.8 to 3.12 and glibc 2.28 or later. Building from source is documented but adds real work. There is no graphical assistant: everything is driven from Python. Onboarding is well signposted, with a five-minute quick start and a two-hour walkthrough, but production deployment is left to the user.
Deployment
Integrations
Supported languages
Behind getML
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement getML.
Frequently asked questions
Who publishes getML?
What kind of data does getML work on?
Is there a free version?
How much does the Enterprise edition cost?
Which systems can getML be installed on?
Does getML provide an API?
Does my data leave my infrastructure?
Can the Enterprise edition be trialled?
What does getML integrate with?
Does the publisher disclose its subprocessors?
Should you pick getML?
getML is a narrow, technically serious product. It does one thing - automated feature engineering on relational and time-series data - and it does so with proprietary algorithms, a patent filed in 2020 and published benchmarks against featuretools, tsfresh and Prophet rather than marketing claims alone. The architecture is unusual in a market dominated by hosted services: the engine runs on the customer's own machines, so training data never reaches the vendor, and learned features are transpiled into readable SQL that outlives the tool itself. For a data science team fluent in Python and responsible for its own infrastructure, that combination is genuinely attractive. The commercial side is far less legible. There is a real open-source entry point in the Community edition, but the paid door is entirely opaque: no pricing page exists anywhere on the site, no amount appears in any of its 157 indexed pages, and Enterprise support is described only as negotiable. Buyers who need a figure before opening a conversation will not find one. Several documentary gaps deserve attention. The trial terms page on the site is published empty, the real legal documents sit on a third-party Notion domain without a resolved update date, and the privacy statement's own text says it covers a different website. No data processing agreement, certification or product security commitment is published. Freshness signals are also weak: the last blog post dates from December 2024, documentation pages carry autumn 2024 dates, and the advertised Enterprise trial download is still described as not yet available. The verdict is therefore conditional. The technology is credible and the free edition lets anyone verify that for themselves at no cost. The commercial and contractual layer around it needs direct clarification with the publisher before any production commitment.
- Choosing a selection results in a full page refresh.
- Opens in a new window.