DEV Community: Archestack

1 in 6 self-hosted AI apps ships a license GitHub can't classify

Archestack — Fri, 09 Oct 2026 11:40:00 +0000

Quick one if you're about to build something on top of a self-hosted AI app. We track 161 of them, and for most GitHub reads a standard license: Apache-2.0 on 63, MIT on 57.

25 of them (16%) ship a LICENSE file GitHub can't map to a standard license. Usually that means custom terms, a dual license, or a normal license with extra conditions added. Some big names are in that group: n8n, AutoGPT, Dify and Open WebUI.

Read the file before you build on one of these. Running it at home is rarely the issue. Hosting it for other people, or selling something on top of it, is where those terms start to matter.

Copyleft is a small slice. 12 apps use AGPL-3.0 (Firecrawl, Khoj, SillyTavern and others) and 3 use GPL-3.0. AGPL only matters if you change the code and offer it to others over a network; on a home server it changes nothing.

Starter templates go the other way. All 89 starters on our second list are permissive: 70 MIT, 17 Apache-2.0, 2 BSD. Makes sense, they exist to be forked into products.

Every app on the list shows its license as GitHub reports it, so the ones worth a closer read are easy to spot: https://github.com/archestack/best-of-selfhosted-ai

Written with help from AI tools. Numbers from the list's public data, checked on 2026-10-09.

83% vs 47%: self-hosted AI apps keep shipping, starter templates sit still

Archestack — Fri, 09 Oct 2026 11:39:44 +0000

We keep two lists: 161 self-hosted AI apps, and 89 starter templates for building AI products. Same check on both: when was the last commit?

Apps: 134 of 161 (83%) had a commit in the last 30 days. 11 more in the last 90 days, 12 in the last six months, and only 4 older than that.

Starters: 42 of 89 (47%). And 18 of them, one in five, haven't seen a commit in over six months.

It lines up with how each gets used. An app runs every day, people file issues, a dependency breaks, someone pushes a fix. A template gets written for a launch or a tutorial, people fork it, and the original stays where it was. An old template can still be a fine reference; just expect older framework versions when you clone it.

One caveat: both lists already drop anything with no commit in 12 months, so these numbers describe the active set, not the whole ecosystem.

Both lists, with the last commit and what each project needs, start here: https://github.com/archestack/best-of-selfhosted-ai

Written with help from AI tools. Numbers from the lists' public data, checked on 2026-10-09.

Only 10 of 89 AI starter templates work with local models out of the box

Archestack — Fri, 09 Oct 2026 11:35:49 +0000

If you want to build an AI app that runs on your own models, the starter you fork matters. We track 89 open-source starter templates for AI apps, and each entry records what the README says about models and what you need to bring.

54 of 89 list an API key as a requirement. OpenAI is the usual one: 49 templates support it, Anthropic shows up in 20, and a lot of them sit on the Vercel AI SDK (21) or LangGraph/LangChain (19).

10 work with local models out of the box, through Ollama or something similar. morphic, agent-service-toolkit, fragments, ai-town and llm-answer-engine are in that group.

In a cloud-first template, swapping the provider is usually a small edit. If running local is the whole point, though, starting from one of those 10 skips that work.

They lean TypeScript, by the way: 56 of 89, with 19 in Python.

The full list, with what each template has wired in and what you still have to build: https://github.com/archestack/best-of-ai-starters

Written with help from AI tools. Numbers from the list's public data, checked on 2026-10-09.

Most self-hosted AI apps don't need a GPU (113 of 161)

Archestack — Fri, 09 Oct 2026 09:49:18 +0000

If you're putting AI stuff on a home server or a cheap VPS, the GPU question comes up fast. We keep a ranked list of 161 open-source AI apps you can self-host, and each entry records what the project's README says about GPUs. Here's how it shakes out.

113 run without one. 31 say a GPU is optional, 8 need one, and 9 READMEs don't say.

The GPU lives in one place: the model server. In model serving, 14 of 19 projects treat a GPU as optional (Ollama, LocalAI, llama.cpp, vLLM and friends) and 5 need one. Image and video is the other spot, with three trainers that won't run without a card.

Everything else talks to the model over an API. Chat UIs, RAG, gateways, agent platforms, memory, observability, sandboxes: almost all of them run on whatever box you've got. Voice is the split one, 6 of 11 can use a GPU if you have it.

So the easy way to plan it: pick the UI and the tools for any machine, and pick the model server for whatever card you have (or don't have).

What the list can't tell you yet is RAM. Only 11 of 161 READMEs give a minimum, so there's no RAM column. Cost estimates and tested builds aren't there either; they're on the to-do list.

The full list, with what each app needs and where it tends to break, is here if you want to keep it around: https://github.com/archestack/best-of-selfhosted-ai

Written with help from AI tools. The numbers come from the list's public data file and were checked on 2026-10-09.