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.