Noureddine RAMDI / Building private AI workflows with the n8n self-hosted AI starter kit

Created Mon, 04 May 2026 10:23:02 +0000 Modified Sat, 23 May 2026 20:41:27 +0000
940 Words
github-stars docker llm automation ai-workflows self-hosted vector-search

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

Spinning up AI workflows that keep your data local and private is becoming a real need, not just a nice-to-have. The n8n Self-hosted AI Starter Kit offers a way to get a full AI agent stack running in minutes, all encapsulated in Docker Compose profiles tailored for different hardware setups. This approach skips cloud APIs entirely, targeting developers who want control, security, and flexibility without sacrificing speed of iteration.

What the n8n self-hosted AI starter kit does

This starter kit is essentially a curated Docker Compose template that bundles several key components for AI-driven automation and data processing into one deployable stack. At its core is n8n, a low-code automation platform with over 400 integrations, which acts as the workflow orchestrator.

Alongside n8n, it includes Ollama for local large language model (LLM) inference, Qdrant as a vector database for efficient similarity search, and PostgreSQL for reliable, scalable data storage. These pieces are pre-configured to work together seamlessly, with networking and persistent volumes set up by default.

The stack supports multiple hardware profiles to optimize performance depending on your machine: Nvidia GPU, AMD GPU on Linux, Apple Silicon (Mac), and CPU-only setups. This allows developers across platforms to prototype AI agents, document summarization tools, retrieval-augmented generation (RAG) pipelines, and private Slack bots without relying on external APIs.

Its architecture relies heavily on Docker Compose profiles, enabling a flexible yet opinionated setup. The communication between components happens over Docker network bridges, making the stack modular yet integrated. The environment is designed more for development and proof-of-concept than production, with convenience and speed prioritized over hardened security or scaling.

Why the n8n stack stands out technically

The combination of a low-code automation engine with local LLM inference and vector search in one package is relatively rare. Most AI workflows often depend on cloud APIs for LLM calls or manage these components separately. Here, the tradeoff is clear: you get full control and privacy, but you must manage local hardware dependencies and resource allocation.

The code quality in the repository is pragmatic. The Docker Compose files are well-structured and use profiles effectively to reduce complexity for users with different hardware. The .env.example file encourages users to customize secrets, which is good practice for real deployments. While the codebase doesn’t dive deep into custom application logic, it provides a solid orchestration foundation.

One technical limitation is that this stack isn’t focused on production-grade resilience or multi-node scaling. For example, the vector database and PostgreSQL are single instances without clustering. This is perfectly fine for prototyping but would require additional engineering for production use.

Another aspect worth noting is the developer experience (DX) around hardware support. The Nvidia GPU profile requires users to have configured Docker for GPU access, which can be a stumbling block if unfamiliar. AMD GPU support is Linux-specific, and Apple Silicon users have their own profile. This shows awareness of real-world hardware diversity but also surfaces the challenge of maintaining smooth DX across platforms.

Quick start with docker compose profiles

Getting started is straightforward if you have Docker installed. The README provides precise commands for cloning, environment setup, and launching the stack based on your GPU or CPU environment.

# Clone the repository and prepare the environment
git clone https://github.com/n8n-io/self-hosted-ai-starter-kit.git
cd self-hosted-ai-starter-kit
cp .env.example .env # customize this file for your secrets

For Nvidia GPU users:

docker compose --profile gpu-nvidia up

For AMD GPU users on Linux:

docker compose --profile gpu-amd up

For Apple Silicon (Mac) users and CPU-only setups, the README continues with similar instructions (not fully detailed here due to truncation).

This setup launches all services defined in the compose file with optimized settings for your hardware, letting you start building AI workflows quickly with local LLMs and automation.

Verdict: who should use this starter kit?

This project is a solid starting point for developers who want to experiment with private AI workflows without cloud dependencies. If you need to process sensitive documents, build internal AI agents, or prototype RAG pipelines with local data, this kit covers the core components out of the box.

Its main value is in rapid prototyping and exploration rather than production readiness. The tradeoffs around single-instance databases and hardware-specific Docker setup mean you’ll need to extend or adapt it for production scenarios.

If you’re comfortable with Docker and want to understand how local LLM inference can integrate with automation tools like n8n, this repo is worth a look. It’s also a useful reference for bundling complex AI stacks into manageable Docker profiles while maintaining developer accessibility.

Overall, the n8n Self-hosted AI Starter Kit is a practical toolkit for building private AI workflows with control, speed, and flexibility — just don’t expect it to be a plug-and-play production platform yet.

→ GitHub Repo: n8n-io/self-hosted-ai-starter-kit ⭐ 14,753