Atarity/deploy-your-own-saas: a curated index of self-hosted replacements for everyday cloud services

What the repository actually contains

The repository is a single curated README organised by the everyday service it replaces. Each category has a heading in the form Deploy your own followed by the commercial product or function being substituted: VPN, Music streaming service, Photo hosting, eBooks catalog, Google Docs, Netflix, Notes, Bookmarks service, URL shortener. Within a category, a markdown table gives four columns: Name, Description, a star count, and an Updated field rendered as a coloured circle plus a relative age such as 1d, 2h, 2y, 8mo. The star column is populated for GitHub-hosted entries and shows n/a for projects whose canonical home is a website rather than a repository, which is the case for WireGuard, Madsonic, Funkwhale and Readeck. The Updated column is the more useful of the two, because it distinguishes a project that saw a commit in the last hour from one that has been quiet for two years. The README image at the top is a file named scryer.jpg stored in the repository root. The topics attached to the repository are awesome, awesome-list, list, open-source and self-hosted, which is an accurate label: this is an awesome-list, and awesome-lists are editorial artefacts rather than software.

Who the list is for, and the decision it speeds up

The reader this serves is someone who has already decided to stop paying for one specific hosted service and now needs candidate names. That is a narrow but real moment. If you are trying to replace a photo library, the Photo hosting table hands you Immich, PhotoPrism, Lychee, Piwigo, Pinry and Chevereto in one screen, each with a one-line positioning statement. The descriptions are doing real differentiation work in a few cases: PhotoPrism is described as personal photo management with AI classification, Pinry as your own Pinterest with boards and tags, Chevereto as an image hosting script for standing up a hosting website. Those are three different products, not three spellings of the same thing, and the list makes that visible. The same is true in the Notes category, where Memos is a memo hub, CodiMD is realtime collaborative markdown, Standard Notes is end-to-end encrypted with a linked self-hosting tutorial, and Bangle.io is web-only WYSIWYG saving locally in markdown. The list is not for someone who wants a single recommendation. It gives you a set and lets you choose, and it does not rank within a category beyond the incidental ordering of the table.

The freshness column is the only maintenance signal, and it is coarse

The Updated field is the closest thing to a health check in the whole repository. It reads as a build artefact rather than something typed by hand: the relative ages (2h, 4h, 16h, 1d, 3d) are too precise and too varied to be maintained manually across dozens of rows. That implies a generator runs on a schedule and rewrites the tables, which would explain why the primary language is listed as Python despite the repository shipping no application code. The README itself does not document the generator, the schedule, or how the colour thresholds are set, so the exact rule that turns a project red rather than orange cannot be confirmed from the material. What can be confirmed is the signal's resolution. A green dot means recent activity; a red dot with 2y or 4y next to it means the project has been effectively dormant. Lychee shows 4y, COPS shows 2y, Leanote shows 2y, and the docker-openvpn entry shows 2y. Orange sits in between at 8mo or 11mo for Pinry, ONLYOFFICE CE, SeaTable, CodiMD, xBrowserSync, Unmark, Briefkasten and TinyList. Treat the dot as a first filter, not a verdict: a dormant project can still be finished software that does its job, and an active one can be active for reasons unrelated to the feature you need.

What the tables leave out, and why that matters before you install

Four columns is a deliberate budget, and the omissions are the cost. There is no licence column, and the repository's own licence field came back unknown, so nothing in the material tells you whether a given entry is MIT, AGPL, or source-available under terms that restrict commercial use. There is no resource footprint, no indication of whether an entry needs a database, object storage, or a GPU for its AI features. There is no note on whether a project is a single container or a multi-service stack. There is no security posture, no CVE history, no statement about which entries have had a third-party audit. The star counts are present but they measure attention, not fitness, and a high count next to a project does not tell you whether it will still be maintained in two years. The descriptions are also marketing-adjacent in places because they are lifted from the projects themselves: Chevereto is described as powerful and fast, which is the project's own phrasing rather than an editor's assessment. None of this makes the list wrong. It makes it a starting point whose second step is always a visit to the linked repository.

Getting value out of it without running anything

There is nothing to install. The repository has no releases and no published package, so the workflow is: open the README, find the category, read the four columns, then follow the link. The one piece of operational guidance the list does carry is a link rather than an instruction: the Standard Notes row points to an external tutorial on completely self-hosting Standard Notes, which is the only entry in the visible material that acknowledges that deployment is the hard part. For everything else you are on your own once you click through. If you want to use the list as an input to your own tooling, the practical route is to fetch the raw README from the master branch and parse the markdown tables, since the table structure is consistent across categories: a header row, a separator row, then one row per project. That consistency is the repository's most underrated property. It is machine-readable by accident, and the four-column schema is stable enough that a small script can turn it into a comparison sheet without any manual transcription.

The alternative: run a search engine instead of a list

The obvious alternative is the awesome-selfhosted catalogue, which covers the same territory with a much wider net. The difference in approach is editorial versus encyclopaedic. This repository curates by everyday need, so a category like Deploy your own Netflix contains exactly one entry, Jellyfin, because the maintainer judged that one project answers the question. A broader catalogue would list several media servers and leave the choice to you. That single-entry category is the clearest illustration of the design: the list is willing to be short and opinionated about what belongs, at the cost of hiding alternatives that a wider index would surface. The trade runs the other way too. Because this list is small, it is readable in one sitting and the freshness column stays legible. A larger catalogue gives you more candidates and less signal about which ones are alive. Neither is strictly better; they answer different questions. This one answers which projects are worth looking at for this specific job. The wider one answers what exists at all.

Maintenance cost, licence exposure and the honest boundary

The maintenance cost of the list itself is paid by whoever runs the generator, not by you. Your cost is the recurring cost of the services you deploy from it, which the list does not estimate. Licence exposure is the real gap. Because the repository carries no licence information and the tables have no licence column, you cannot tell from this material whether a given entry is safe for the use you have in mind. That is not a legal opinion and should not be read as one; it is a statement that the data needed to form one is absent here and has to be fetched from each linked project. The same applies to upgrade cost. A project showing a green dot is being changed, and changed software needs upgrading, but the list does not tell you whether upgrades are drop-in or require migrations. The one category where the list hints at operational weight is Google Docs, where ONLYOFFICE CE is noted as distributed as a Docker image with a link to the installation documentation, and Collabora is described as LibreOffice in the cloud. Those are heavier deployments than a bookmark manager, and the table does not say so. The list is a map of names. The terrain is in the repositories it points to.

Editorial conclusion

Use this repository as a shortlist generator when you have already decided to move a specific service off a hosted provider and need candidate names for that one category. Do not use it as a deployment guide, a licence check, or a security review; the tables contain none of that. Before you install anything from it, open the linked project's own repository and confirm its licence, its release cadence and its backup story, because the only signal this list gives you is a name, a one-line description and a colour-coded freshness dot.

Official sources

1. Atarity/deploy-your-own-saas on GitHub
2. Issues
3. README