n8n-io/self-hosted-ai-starter-kit

The Self-hosted AI Starter Kit is an open-source template put together by the n8n team. The goal is to give you a ready-made local setup for building AI workflows on your own machine instead of relying on a hosted cloud service. Everything is wired together with Docker Compose, so once you have Docker installed you run a single command and the whole stack comes up. The kit bundles four pieces. The first is n8n itself, a low-code automation platform with over 400 integrations and nodes for AI work such as agents, text classifiers, and information extractors. The second is Ollama, a tool for downloading and running local large language models. The third is Qdrant, an open-source vector database for storing the numeric embeddings that AI applications use to look things up by meaning. The fourth is PostgreSQL, used as a general-purpose database. The README lists possible projects you might build with this combination: scheduling agents, secure summarisation of company PDFs, smarter Slack bots, and private analysis of financial documents. Installation comes in a few flavours. You clone the repository, copy the example environment file, and then choose a docker compose profile that matches your hardware. There are separate profiles for NVIDIA GPUs, AMD GPUs on Linux, and plain CPU. On Apple Silicon Macs the GPU cannot be passed into Docker, so the README offers two routes: run everything on CPU, or install Ollama on the Mac directly and have the Docker stack talk to it through host.docker.internal. Once the stack is running you open n8n at localhost port 5678 to do a one-time setup, then load the example workflow and click the Chat button to try it. The first run downloads the Llama 3.2 model through Ollama, which may take a while. After that, the README suggests keeping things local by always picking the Ollama node for language models and Qdrant for vector storage. The README closes with upgrade commands for each hardware profile, a list of recommended reading on AI agents and vector databases, a video walkthrough, and links to the n8n template gallery so you can import other workflows. It notes that the kit is aimed at proof-of-concept work and is not tuned for production deployments.