Fe/male Switch: Your Startup Facilitator & Incubator for Women

Open-Source Product Recommendation Engines (2027): 10 Self-Hosted eRecommender-Style Tools for Online Shops

Open-Source Product Recommendation Engines for Online Shops and Startups

"Customers who bought this also bought", "you may also like", "recommended for you": product suggestions like these are what people mean by an eRecommender. We looked for an official product called eRecommender and could not find one: there is no website, app listing or code repository under that name, only an old university thesis and similarly named commercial services. So this guide answers the real question behind the search: which open-source recommendation engine can a small online shop or startup host itself, without paying a per-order or per-visitor fee to a recommendation vendor?
A lot has changed since the 2025 version of this list. Several projects on it, such as Rexy, Spotlight and TensorRec, are abandoned, and Apache Mahout has moved to quantum computing. We checked the licence, latest release and recent activity of every project below on its GitHub repository, PyPI page or official website in October 2026.
Quick answer: for an online shop that wants working recommendations without a data science team, start with Gorse: a ready-made service with a REST API, a dashboard and Docker images (Apache-2.0). Use Metarank to personalise the ranking of results you already have, and Vespa if you need search and recommendations together at scale. If you have a Python developer, Implicit builds strong "also bought" models from clicks and purchases, and Microsoft's former Recommenders project shows how to do it step by step.
Boost Your SEO by Getting Featured in Our Blogs and get a backlink.

We publish content about startups, education, tech, funding, etc. that ranks well not only in Google but also in Perplexity, ChatGPT, Grok and other AI tools.

👉 Get featured now!

Open-Source Recommendation Engines Compared

ProjectWhat it isLicenceLanguageLatest release (checked Oct 2026)
GorseRecommender service with REST API and dashboardApache-2.0Gov0.5.11, July 2026
MetarankPersonalised ranking serviceApache-2.0Scala0.8.1, September 2026
VespaSearch, ranking and recommendation engineApache-2.0Java, C++v8.763.13, October 2026
RecommendersBest-practice notebooks and algorithmsMITPython1.2.1, December 2024 (code still updated)
ImplicitCollaborative filtering for clicks and purchasesMITPython0.7.3, May 2026
CornacFramework for multimodal recommendersApache-2.0Python3.0.1, September 2026
LensKitToolkit for training and evaluating recommendersMITPython2026.4.0, September 2026
TensorFlow RecommendersLibrary for deep recommendation modelsApache-2.0Python0.7.7, January 2026
TorchRecPyTorch library for large-scale recommendersBSD-3-ClausePython1.9.2, October 2026
SurpriseLibrary for explicit ratingsBSD-3-ClausePython1.1.5, May 2026
Sources: each project's GitHub repository and releases, PyPI page and official website, checked 27 October 2026.

Engine, Library or Search Platform: What Do You Need?

The projects on this list solve the same problem at three very different levels of effort:
  • A ready recommendation service (Gorse, Metarank): you run it with Docker, send it your products and what shoppers do, and ask its API for recommendations. A developer who can build a web shop can set it up; no machine learning knowledge is needed to start.
  • A search and ranking platform (Vespa): one system for product search, filters and recommendations. It is powerful but takes real engineering time to model your data and ranking.
  • A Python library (Implicit, Cornac, LensKit, Recommenders, Surprise, TensorFlow Recommenders, TorchRec): you train models in your own code and build the serving layer yourself, usually by computing recommendations in a nightly job and storing them in your database.
All of them are free to use. The cost is hosting and your team's time, so a small shop should begin with a service and only move to a library when it has the data and the people to beat it.

The 10 Best Open-Source Product Recommendation Engines in 2027

1. Gorse: a ready recommender service for online shops

Gorse is an open-source recommender system written in Go and released under Apache-2.0. You import items, users and feedback (for example views, likes or purchases), and Gorse trains its models automatically and serves recommendations through a REST API. It combines several sources: latest items, item-to-item, user-to-user and collaborative filtering, and it supports multimodal content (text, images, video) through embeddings as well as LLM-based recommenders.
It stores data in MySQL (or MariaDB), Postgres or MongoDB, with Redis as a cache option, and runs as a binary, in Docker or on Kubernetes. There are client SDKs for Go, Python, TypeScript, Java, Rust, PHP, Ruby and .NET, and a GUI dashboard for editing the recommendation pipeline, monitoring and online evaluation. A playground mode starts a working demo with one Docker command. The latest release is v0.5.11 (July 2026), and the code is still updated (last commit in October 2026).

2. Metarank: personalised ranking as a service

Metarank is an Apache-2.0 ranking service written in Scala. Instead of replacing your shop's search or product lists, it re-orders them for each visitor using signals such as clicks, purchases, CTR and the visitor's session, and its README says it can re-rank large result sets within 10 to 20 milliseconds. It also offers trending and similar-item recommendations, and works next to Elasticsearch or OpenSearch. Releases continue (0.8.1 in September 2026), although most recent commits are automated dependency updates, and the metarank.ai website did not respond when we checked; the documentation is online.

3. Vespa: search and recommendations in one platform

Vespa is an open-source engine (Apache-2.0) for search, ranking and machine-learned inference over text, vectors and structured data, and its site names recommendation and personalisation as a core use case, including e-commerce, where search and recommendations need to work together. All the code needed to run it yourself is on GitHub, and it is actively developed (v8.763.13 came out in October 2026). Vespa Cloud is the company's paid hosted version; its pricing page offers a free trial with $300 in credits. Vespa is the most powerful option here, and also the heaviest: choose it when product search and recommendations are central to your business.

4. Recommenders: best practices and algorithms in Python

Recommenders started at Microsoft and is now a project of the Linux Foundation AI and Data, under the MIT licence. It is a collection of Jupyter notebooks and Python utilities that walk through the five tasks of a real project: preparing data, building models (from ALS to deep models such as xDeepFM), evaluating them offline, tuning them and putting them into production. The last PyPI release is 1.2.1 from December 2024, but the repository is still updated (last commit in October 2026). It is the best place to learn how recommendation systems work before you build one.

5. Implicit: fast "also bought" models from clicks and purchases

Implicit (MIT) is a Python library for implicit feedback: the views, clicks and purchases that every shop has, rather than star ratings. It implements Alternating Least Squares, Bayesian Personalized Ranking, Logistic Matrix Factorization and item-to-item nearest-neighbour models, trains on all CPU cores and can use an NVIDIA GPU. It installs with pip on Linux, Windows and macOS, and a few lines of code give you recommendations per user and similar items per product. Version 0.7.3 came out in May 2026.

6. Cornac: recommenders that use product text and images

Cornac is an Apache-2.0 Python framework from Preferred.AI for multimodal recommender systems: models that use extra data such as product descriptions, images and social links, which can help with new products that have little sales history. It also makes it easy to compare many models on the same data with standard metrics. Version 3.0.1 came out in September 2026.

7. LensKit: test and evaluate recommendation algorithms

LensKit is a set of Python tools for training, running and evaluating recommender algorithms, built for research and education and published under the MIT licence. It is useful when you want to measure which algorithm works best on your own sales data before you build anything. It is actively developed, with release 2026.4.0 in September 2026.

8. TensorFlow Recommenders: deep learning models with Keras

TensorFlow Recommenders (TFRS) is an Apache-2.0 library for building recommender models, from data preparation and training to evaluation and deployment. It is built on Keras and supports retrieval models that use user, item and context features, plus multi-task models. Version 0.7.7 came out in January 2026, and PyPI still labels the package as alpha. Choose it if your team already uses TensorFlow.

9. TorchRec: large-scale recommenders on PyTorch

TorchRec is a BSD-licensed PyTorch library for large recommendation models whose embedding tables are split across many GPUs; its README says it powers many production recommendation models at Meta. Stable version 1.9.2 came out in October 2026. It is far more than a small shop needs, but worth knowing if your startup builds recommendation technology itself.

10. Surprise: a simple library for star ratings

Surprise is a BSD-licensed Python library for building and testing recommenders on explicit rating data, with classic algorithms such as SVD, SVD++, NMF and neighbourhood methods and easy cross-validation. Its website states that it does not support implicit ratings or content-based information, so it fits shops with product reviews or star ratings, not click data. Version 1.1.5 came out in May 2026.
Validate your startup idea with the unique borrowed authority approach: we publish articles about your product in our blog and you get traffic and testers for your MVP

  • Prove Market Demand: See real organic traffic and waitlist conversions

  • Unlock High-Potential Keywords: Receive a curated list of top-performing keywords directly from Google Search Console data.

  • Estimate Customer Acquisition Cost (CAC): Gain financial foresight with an estimated CAC based on real keyword performance data.

🔗 Start validating your startup now

Recommendation Engines That Are Discontinued or Slowing Down

Several projects from older lists, including the 2025 version of this one, are no longer a safe choice for a new shop:
  • Apache PredictionIO: retired in September 2020 and moved to the Apache Attic in April 2021; its repository is archived.
  • Universal Recommender and Harness (ActionML): the Universal Recommender moved from PredictionIO into the Harness server; Harness's last release dates from March 2021 and its last commit from September 2021.
  • Apache Mahout: the project now focuses on quantum machine learning (Qumat and QDP), and the classic Mahout code with its MapReduce algorithms is in maintenance mode.
  • TensorRec: its README says it is not under active development; the last release dates from April 2019.
  • Spotlight: last release in September 2019 and last code change in February 2020.
  • Rexy: its GitHub repository no longer exists.
  • Slowing down: LightFM (last release March 2023, no code changes since April 2023), RecBole (last release and commit in February 2025) and NVIDIA Merlin (last tagged release June 2024) are still available, but check their activity before you depend on them.

Which Open-Source Recommendation Engine Should You Choose?

  • You run a small online shop and want recommendations this month: Gorse.
  • You already have search and product lists and want them personalised: Metarank.
  • Search and recommendations are your core product: Vespa.
  • You have a Python developer and purchase or click data: Implicit, with Recommenders or LensKit to compare models.
  • Many new products with little sales history: Cornac, using product text and images.
  • Your team works in TensorFlow or PyTorch: TensorFlow Recommenders or TorchRec.
  • You only have star ratings: Surprise.
Prefer a free or paid tool instead of hosting your own? See our lists of free eRecommender alternatives and eRecommender alternatives. If you are still choosing the shop itself, compare open-source Shopify alternatives.

How Founders Use Product Recommendations

Recommendations only work on good data, so most of the job happens before the engine: tracking what shoppers view, add to the cart and buy. Our guides to open-source Segment alternatives and open-source Mixpanel alternatives cover tools that collect those events. Recommendations also work outside the shop page: e-commerce brands send them by email, and our Omnisend vs Klaviyo comparison shows which email tools support product blocks. Good product pages matter too; see our free product description generators.
If you are building a recommendation feature into your own product, start small: our guides on building an AI MVP on a budget and building an AI-powered app from scratch show how, and how to grow your customer base explains where recommendations fit into growth. To meet other online retail founders, see the top e-commerce startup events.
Open source is a good default for the rest of a shop's stack as well: see our guides to open-source Intercom alternatives for live chat, open-source Freshdesk alternatives for customer support, open-source Buffer alternatives for social media, open-source Expensify alternatives for expenses and open-source GitHub Copilot alternatives for the developers who build it all. To get your shop found in Google and AI answers, try our AI SEO tool.

FAQ About Open-Source Product Recommendation Engines

What is eRecommender?

We could not find an official product called eRecommender. The name is best read as a description of product suggestion apps for online shops: "also bought", "you may also like" and personalised product blocks. This guide lists open-source engines that do the same job.

What is the best open-source recommendation engine for an online shop?

For most small shops, Gorse: it is a ready service (Apache-2.0) with a REST API, a dashboard and Docker images, and it trains its models automatically from your product and shopper data.

Can I add product recommendations without a data scientist?

Yes, with a service such as Gorse. You send it products and events (views, add-to-cart, purchases) and ask its API for recommendations. A web developer can set it up; tuning it further is where machine learning knowledge helps.

What data do I need for product recommendations?

A product catalogue (with categories, tags or descriptions) and shopper events such as views, cart additions and purchases, each with a user, a product and a time. Star ratings help but are not required.

Is Apache PredictionIO still maintained?

No. PredictionIO was retired in September 2020 and moved to the Apache Attic in April 2021. Its repository is archived.

Is Apache Mahout still a recommendation engine?

Not in practice. Mahout now focuses on quantum machine learning (Qumat and QDP), and the classic Mahout code with its MapReduce algorithms is in maintenance mode.

Is LightFM still maintained?

LightFM is still on PyPI, but its last release, 1.17, came out in March 2023 and the code has not changed since April 2023, so check its activity before you depend on it.

Which library is best for clicks and purchases rather than ratings?

Implicit (MIT) is built for implicit feedback such as clicks and purchases, with ALS, BPR and item-to-item models. Surprise only supports explicit ratings.

Are open-source recommendation engines free?

The software is free under licences such as Apache-2.0, MIT and BSD. You pay for hosting and for the time to set it up, and hosted versions such as Vespa Cloud are paid.

How do I know if recommendations will sell more in my shop?

Test it: show the recommendation block to part of your visitors and compare clicks and orders. Before you build anything, you can also test demand for your product idea with our startup idea validation service.

About the Author

Violetta Bonenkamp, also known as MeanCEO, is an experienced startup founder with an impressive educational background including an MBA and four other higher education degrees. She has over 20 years of work experience across multiple countries, including 5 years as a solopreneur and serial entrepreneur. Throughout her startup experience she has applied for multiple startup grants at the EU level, in the Netherlands and Malta, and her startups received quite a few of those. She’s been living, studying and working in many countries around the globe and her extensive multicultural experience has influenced her immensely.
2026-10-27 01:51 Top Alternatives