The annual self-hosting survey opens with a first AI section and 45 questions through October 31
selfh.st’s 2026 user survey opened on October 7, 2026 with 45 questions and, for the first time, a section dedicated to AI. If you run self-hosted services, respond before the October 31 deadline: it is the largest public dataset on the community, and its results will shape the ecosystem in 2027.
October 7, 2026. Ethan Sholly, the creator of selfh.st, opened the community’s 2026 user survey. This year the form has 45 questions and, notably, a section dedicated to AI. Entries close on Saturday, October 31 at 9 p.m. ET, and the results will be published in early November. Why it matters: this is the only large-scale annual census of what self-hosters actually run, and the arrival of an AI section signals that self-hosted AI has reached critical mass.
What the survey measures
The 2026 survey follows the line of its predecessors: hardware, operating systems, deployed applications, storage, backups, and networking. The new part is the AI section, which tries to quantify a shift the rest of the form could not capture: how many self-hosters run a local LLM, an image-generation service, or an assistant on their own hardware rather than in the cloud.
The choice of topic is not incidental. Over the past two years, self-hostable AI tooling — Ollama, LocalAI, interfaces like Open WebUI — has moved from a curiosity to a first-class homelab workload, on par with a media server or a feed reader. The survey is the first instrument that can measure that adoption across the community, instead of relying on forum anecdotes.
The AI section also sits at the intersection of two questions the rest of the form asks separately: what hardware you run and what you actually deploy. A self-hoster who reports a GPU in the hardware section but no AI in the AI section is itself a signal — it may mean the GPU sits idle, or is used for transcoding, or that the user tried local AI and gave up. Cross-referencing those answers is where the survey’s real insight will come from.
For a community that has historically measured itself in apps and hardware, adding AI to the census is an admission that the definition of “self-hosting” is expanding. Whether that expansion sticks will be answered, in numbers, when the results land in November.
The community’s most reliable census
The exercise’s credibility rests on its volume and its method. The 2025 edition collected 4,081 completed responses, captured with Formbricks — an open-source, self-hosted survey tool that also sponsored the edition — and visualized with Chart.js, with the raw data published on GitHub. In short: the self-hosting community’s census is itself self-hosted, and its results are reproducible, which sets it apart from closed polls run by commercial vendors.
That transparency has a concrete consequence for the ecosystem. The survey’s results serve as a reference for maintainers choosing their priorities, for writers deciding what to cover, and for developers gauging appetite for a new application category. When a tool climbs the rankings year over year, it gains visibility — and therefore contributors and funding. Responding is a way to weigh on that signal.
The survey’s value also comes from its continuity. Each edition reuses the same blocks — hardware, OS, applications, storage — which makes it possible to track year-over-year trends: an app that rises, a distribution that falls, a deployment method that takes over. Those long series, more than any snapshot, are what make the dataset usable by maintainers. A tool that climbs three years running is not a fad: it is a trajectory.
With 4,081 completed responses in 2025, the margin of error is low for a community poll, and the sample spans a wide spectrum — from the apartment server to the enterprise homelab. It is not a random sample of the general population, but it is precisely the population that matters to the ecosystem: the people who install, maintain, and fund self-hosted software.
Publishing the raw data on GitHub has a direct consequence: anyone can re-verify the aggregates, cross the variables, or spot angles the official report did not surface. That is rare in the polling world, where data usually stays the author’s property, and it is what makes the survey a dataset as much as a report.
Why the AI section is a turning point
The addition of an AI section is the clearest sign that self-hosting is entering a second phase. The first — infrastructure services such as NAS, reverse proxies, monitoring, and backups — is mature and well documented. The second, local AI, runs on different hardware: GPUs, video memory, cooling, power draw. Those are constraints a classic questionnaire does not capture, and only a dedicated section can measure.
The risk for the community is misreading the signal. If the AI section reveals heavy adoption on small hardware — mini-PCs, ARM boards — it will push developers toward quantized, frugal models. If it instead reveals a concentration on GPU-equipped servers, it will validate the dedicated-accelerator path. Either way, the data will be more useful than an opinion debate, and that is exactly what the survey sets out to produce.
The blurry line between “vibe coded” and “AI-assisted”
The community is already wrestling with this ambiguity. In his October 2, 2026 newsletter, Ethan Sholly described the difficulty of tagging “vibe coded” projects: most projects use AI in some capacity, and too many readers were mistaking “AI-assisted” — an author delegating part of the code to a model — for “vibe coded” — a tool produced almost entirely by AI. The AI-assisted tags had in fact been dropped from the newsletter a few weeks earlier, for lack of a clean way to classify every submission.
That vocabulary dispute sheds light on what the survey’s AI section will have to settle. Asking “do you self-host AI” is one thing; asking “which of your self-hosted services were built with AI” is another — and it is the question the community has not been able to answer through editorial alone. The survey has a chance to supply the missing data: not an opinion about vibe coding, but a measure of its real penetration into self-hosted stacks.
How to prepare your answer in ten minutes
The form covers hardware, operating system, applications, storage, backups, networking, and now AI. Before opening the questionnaire, a quick inventory is enough to make the answer precise: list your deployed applications (with their deployment method — Docker Compose, Kubernetes, bare metal), note your hardware (model, amount of RAM, whether a GPU is present), and spot anything that touches local AI — an LLM via Ollama, an Open WebUI interface, an image-generation service.
That inventory is worth doing beyond the survey: it is the same exercise as a backup audit or a hardware review that every self-hoster ends up doing sooner or later. Answering with exact data rather than from memory also makes the dataset more reliable for those who use it — which is exactly what the community expects from a census it imposes on itself.
Verdict
If you self-host even one service, respond before October 31: ten minutes is enough, and every response feeds a dataset that maintainers and publishers rely on all year. If you are still undecided about local AI, wait for the results in early November: the AI section will tell you, with numbers, whether your use case is marginal or already mainstream in the community. Either way, do not skip the one annual appointment that, more than any article or forum, describes what self-hosting is actually becoming.