Free, verified, CC-BY-4.0
The open alternatives dataset
Every verified open-source and self-hostable alternative we track, with the licence read from the project’s own repository rather than its marketing page. 393 entries across 78 categories and 86 proprietary products. Free to download, free to use commercially, and free to check our work against.
- Alternatives
- 393
- Open source
- 299
- Self-hostable
- 314
- Categories
- 78
- Products covered
- 86
What is in it
| field | what it holds |
|---|---|
| alternative | Name and slug of the open or self-hostable option |
| replaces | The proprietary product it substitutes for, and its vendor |
| category / stack_layer | Category, plus which of the five stack layers it sits at |
| license | SPDX-style label read from the project's own LICENSE file |
| open_source | True only for genuinely open licences — qualified ones are false |
| self_hostable / local_first | Whether you can run it, and whether it works offline |
| sovereignty_score | 0–100 control measure; weighting published in the methodology |
| pricing | Pricing shape in plain words, as stated by the project |
| website / repo / docs | Canonical links |
| last_verified | The date that row was last checked |
Coverage by stack layer
- L1Energy0 products · 0 alternatives
- L2Chips & silicon0 products · 0 alternatives
- L2.5Compute0 products · 0 alternatives
- L3Infrastructure29 products · 135 alternatives
- L4Models & tooling14 products · 70 alternatives
- L5Applications42 products · 183 alternatives
Five more datasets, by request
The catalogue above is open to everyone. These five go deeper and are free for research and evaluation (CC-BY-NC-4.0): four short lines on the request form and the download starts at once — no account, no waiting. Each is built from the live site at the moment you download it.
- The free-tier index 80 rows · one per product · from /pricing
Is it free? For every product we cover: the verdict, the exact limits, where the free tier stops, what paid costs from, and the vendor pages it was read from — each dated.
- The head-to-head table 695 rows · one per pair of tools · from /compare
Every comparison on the site as one table: both tools' licence, Sovereignty Score, open-source, self-host and local-first facts, pricing, and — where we have run them — real RAM, monthly cost, setup time and upkeep.
- The self-hosting cost table 169 rows · one per tool · from /run
Documented minimum RAM against what each tool really needs, CPU, disk, realistic monthly cost, setup time, upkeep, the one thing that bites, and when not to self-host it.
- The AI accelerator table 26 rows · one per accelerator · from /silicon
Memory, bandwidth, compute, power, interconnect, software stack, availability, who each part is for and the catch.
- The power and cooling table 23 rows · one per option · from /energy
Lead time, capex, carbon, maturity, density supported, PUE, the catch and what each option decides.
Prefer Hugging Face? The accelerator and power tables are also there as gated sets: macrostack-silicon and macrostack-energy. The copies on our own site are always the newer ones.
Commercial licences
Using one of the request-only tables inside a product, a client deliverable or anything you sell needs a written licence — for AI and search products grounding answers in current pricing and licence facts, analysts and consultancies, and vendors tracking their own category. A licence covers the use, refreshed files as the site is re-verified, and the history of what changed. Terms are written per use; tell us what you are building and a person replies within two business days, with the data yours to evaluate meanwhile.
Using it — attribution
Licensed CC-BY-4.0. Commercial use, redistribution and modification are all permitted. The one condition is credit with a link. Copy whichever form fits:
Plain text
Macrostack (2026). Macrostack Open Alternatives Dataset (version 2026-10-04). CC-BY-4.0. https://www.macrostack.net/data
HTML
<a href="https://www.macrostack.net">Data from Macrostack</a> — CC-BY-4.0
BibTeX
@misc{macrostack2026,
title = {Macrostack Open Alternatives Dataset},
author = {Macrostack},
year = {2026},
note = {Version 2026-10-04. CC-BY-4.0},
url = {https://www.macrostack.net/data}
}Common questions
- Can I use this dataset commercially?
- Yes. It is published under CC-BY-4.0, which permits commercial use, redistribution and modification. The single condition is attribution: credit Macrostack and include a link to macrostack.net wherever you use it. That applies whether you are publishing an article, building a product, or training a model on it.
- How are the licences verified?
- By reading the project's own LICENSE file or repository metadata, not the marketing page. This matters more than it sounds: several widely-described 'open source' projects ship under licences that are not open source at all — Apache-2.0 with additional conditions, community licences with usage thresholds, or GPL with a commercial revenue clause. Those are recorded as what they are, and flagged as not open source, because the distinction is the entire point.
- How often is it updated?
- It regenerates on every deploy from the same catalog the site itself renders from, so the download can never drift from what the pages say. Each row carries its own last_verified date, so you can filter to entries checked within any window you consider acceptable.
- What is the Sovereignty Score?
- A transparent 0-100 measure of how much control an option gives you: licence openness, whether it can be self-hosted, whether it works without a network, and how portable your data is. The full weighting is published on the methodology page. It is editorial, it is consistent, and it is never influenced by whether an entry has an affiliate programme.
- Why publish this for free?
- Because a catalog that nobody can check is a claim rather than a resource. Publishing it in full lets anyone verify our rankings against the underlying facts, which is the only way an independent comparison site earns trust. It also means researchers and journalists can cite the data directly rather than paraphrasing a page.