The AI Stack That Runs Itself: Six Projects for Building, Hosting, and Automating with LLMs
From running open models on your own hardware to wiring agents into real workflows, these six open-source projects cover the full path from local inference to production automation. Each one is actively maintained and solves a distinct layer of the AI stack.
Living article: rankings and health notes are re-checked against GitHub and community data. Ratings are GitPalace community ratings; stars are GitHub stars.
Ollama is a single-binary runtime for downloading and serving open-weight language models locally on macOS, Windows, Linux, or Docker. It exposes a simple local API with official Python and JavaScript libraries, so applications can talk to a model on your machine the same way they would talk to a hosted provider. It is the natural first install for developers who want private, offline inference without wrangling GPU toolchains by hand.
Open WebUI is a self-hosted, provider-agnostic chat interface that works with Ollama and any OpenAI-compatible API. It layers document upload, retrieval, user accounts, and extensibility on top of raw model endpoints, making local models usable by a whole household or team. It is aimed at anyone who wants a polished ChatGPT-style experience that runs entirely on their own infrastructure.
n8n is a fair-code workflow automation platform with a visual canvas, native AI agent nodes, and the ability to drop in custom code when a node falls short. It connects to a very large catalog of third-party services and can be self-hosted or run in the cloud. It suits teams who want to operationalize LLM calls inside real business processes rather than in one-off scripts.
Dify is an open-source LLM application development platform that unifies agent workflows, RAG pipelines, prompt management, model routing, and observability in one collaborative workspace. It gives product teams a shared place to design, test, and ship AI features without each engineer reinventing the plumbing. It fits organizations building customer-facing LLM apps who want visibility into how those apps behave.
LangChain is the framework most developers reach for when composing LLM calls, tools, retrievers, and memory into agents in code. Its strength is an enormous ecosystem of integrations and a higher-level Deep Agents package for getting started quickly. It is best for engineers who want programmatic control and are comfortable living in Python rather than a visual canvas.
Browser Use lets an AI agent drive a real web browser the way a person would: opening pages, clicking, typing, and filling forms from a plain-language task description. It is the missing piece for automating websites that have no API, from data extraction to form submission. It is aimed at developers building agents that must act on the web rather than only read from it.
