Enabling Technologies
Infrastructure
None of this would be as realistic without the current moment. Self-hosting used to mean racking a server for a specialist audience. VPS providers and proprietary cloud platforms made the next step easier — renting compute instead of owning hardware — but still through a closed dashboard, on a closed platform. Now public cloud infrastructure like OpenStack turns provisioning into a few API calls on an open standard, and AI-assisted coding makes maintaining the glue code far more accessible. What used to need a dedicated sysadmin is within reach of a small, motivated group of members.
That doesn't have to mean defaulting to a hyperscaler, either. Steward-owned, low-margin green providers — Infomaniak, Greenhost, ProcoliX — offer the same OpenStack-grade infrastructure without AWS's ownership structure or footprint. They exist because a lot of people did a lot of work, over a long time, to distribute power and make hosting greener — that effort is worth relying on. Good providers also reduce a lot of the harm data centres cause in the first place, like Infomaniak's own D4 Datacenter: modest enough in size to disappear underground within a neighbourhood, and to heat it in the process. See Providers for how they compare.
FOSS
Self-hosting on its own only solves half the problem. Proprietary software on your own hardware still leaves you dependent on a vendor's roadmap and continued existence — you've just moved the rented server into your own building. FOSS — Free and Open Source Software — is what makes self-hosting collectively and publicly owned: the source can be audited, modified, and forked by anyone, and even an abandoned project keeps running rather than leaving you with a dead service.
None of that exists by accident. It's decades of work by a FOSS community — Linux, Nextcloud, WordPress, Forgejo/Codeberg, and many others — building and maintaining software in the open, for free, for anyone to use. Some of that work is explicitly activist: Disroot and Riseup build and run infrastructure specifically for movements and organisers, not just general users. That effort is worth relying on too.
AI
AI makes it trivial to write custom software — but you need to watch out. It's a quiet route back to the same problem. Code written for one organisation, kept closed, and understood by nobody outside it is private and proprietary in every way that counts, whether or not a vendor sold it or a licence fee was paid. Treating FOSS as the default is what keeps AI-assisted development from re-privatising the commons it could otherwise help build. Contributing back would be even better, but using it instead of reinventing it closed is the least we can do.
There's an economic and environmental concern here too. Every organisation regenerating its own version of the same tool from scratch duplicates compute for no good reason — tokenmaxxing a bespoke solution into existence has a real cost and ecological footprint. That cost doesn't stay with the company generating it either: it's externalised onto the commons and civic infrastructure, as rising hardware and energy costs, that everyone else then has to absorb — Codeberg has documented what this looks like from the receiving end. This is AI companies reproducing the need for themselves: for every change, you'd need them again — and pay for them again. One shared, well-maintained FOSS project beats a hundred AI-generated reinventions of the same wheel, in line with the socio-economic commitments this stack already makes.
The open AI community is much younger than FOSS, but it's alive and kicking. Open-weight models are part of it, and so is the tooling around them. Open WebUI is a self-hosted chat interface for running them. Coding agents like OpenCode do the same for development work. Search matters too — a model is only as good as what it can look up, and initiatives like Staan keep that open as well. And memory for agents matters the way documentation does: without it, every session starts from zero. It's a less settled ecosystem, but the same logic applies: rely on and contribute to it rather than defaulting to a closed API.
None of this is a clean story, though. The open AI community is a contested one — it rests on models trained on other people's work without consent, a deep violation of rights that no amount of open-weighting afterwards undoes. And LLM use is corroding the FOSS commons it draws on, undermining trust by driving up the transaction cost of collaboration: Codeberg describes maintainers buried under low-effort, LLM-generated contributions, copyleft code "license-laundered" of its reciprocity requirements by being generated out of training data, and contributors accusing each other of using LLMs when they didn't. The result is a vicious cycle: collaboration gets less rewarding, so people vibecode a one-off tool for themselves instead of contributing to a shared one.
But not using AI at all isn't the answer either: it leaves commercial enterprise free to commoditise the commons while the commons can't defend itself with the same tool. Not every commons is as forgiving as open source, though — its permissive side already legitimises reselling code as proprietary, an accommodation most commons lack. Organisations that aim to democratise, decentralise, definancialise and decommoditise their domain use AI to apply FLOSS, not to whitewash code and commit it as something it isn't. This stack only draws on AI within that context. In the end, using AI to defend the commons carries the same debate as any tool with an unclean lineage: not a clean win, just a battle not fought today. Fighting every battle at once paralyses you and leaves you defenceless.