TutorialProvider guide
Prototype local AI workflows with n8n, Ollama and Qdrant
Docker Compose runs n8n with PostgreSQL, Ollama for inference and Qdrant for vectors, and imports a sample workflow.
Read the source Open the repository
Before you start
- Git, and Docker with Docker Compose.
- For a GPU profile, an NVIDIA GPU already set up for Docker by Ollama's Docker instructions, or an AMD GPU on Linux.
- On a Mac with Apple silicon the GPU is not exposed to Docker: run the kit on the CPU, or run Ollama on the Mac and point the kit at it.
Steps
- Clone the self-hosted-ai-starter-kit repository, enter its folder, copy .env.example to .env and change the secrets and passwords in it.
- Start it with docker compose --profile cpu up, or with the gpu-nvidia or gpu-amd profile; on a Mac, with plain docker compose up.
- Open http://localhost:5678/ and set up n8n. This happens once.
- Open the included workflow at http://localhost:5678/workflow/srOnR8PAY3u4RSwb and press Chat at the bottom of the canvas.
What you get
n8n runs on your machine with Ollama, Qdrant and PostgreSQL, and the included chat workflow answers once Ollama has downloaded Llama3.2, which the Docker logs show. The kit says it is built for proofs of concept and is not fully optimized for production.
Dossiers
- Self-hosted AI starter kitthis guide's dossier
- n8na tool it uses
- Ollamaa tool it uses
- Qdranta tool it uses