What Is the n8n Self-Hosted AI Starter Kit?
The Self-hosted AI Starter Kit is a Docker Compose template, not a new piece of software — its own GitHub description calls it "an open, docker compose template" that helps you "swiftly initialize a comprehensive local AI and low-code development environment." It combines four separately maintained open-source projects (n8n, Ollama, Qdrant, and PostgreSQL) into one pre-wired configuration, so you get a working local AI stack without manually installing and networking each piece yourself.
- Core function: a
docker-compose.ymland supporting configuration files that start n8n, Ollama, Qdrant, and PostgreSQL together, pre-networked so they can talk to each other - n8n: a low-code workflow automation platform with over 400 integrations and dedicated AI/LangChain nodes for building agents and automations
- Ollama: runs open-weight language models locally, exposed to n8n workflows as a model provider
- Qdrant: an open-source vector database used for storing and retrieving embeddings in RAG-style workflows
- PostgreSQL: the relational database backing n8n's own workflow and credential storage
- Maintainer: n8n, the company behind the n8n workflow automation platform
- Canonical repository: github.com/n8n-io/self-hosted-ai-starter-kit — the source of the install steps and version history used in this review
The Starter Kit's Project History and Recent Changes
The kit has no versioned releases; its commit log is the only change history. The GitHub repository publishes no releases or tags, so the dated commits below, read from the default branch, are what can be verified.
- 1Core four-service bundle
Why it matters: The compose file wires n8n, Ollama, Qdrant and PostgreSQL together, imports a demo workflow and two pre-made credentials ("Local Ollama service" and "Local QdrantApi database") on first start, and pulls the `llama3.2` model automatically. - 2GPU Compose profiles
Why it matters: Separate `gpu-nvidia` and `gpu-amd` profiles sit alongside the default `cpu` profile, so Ollama can use a supported GPU instead of running every model on the CPU. - 3September 1, 2025 — Demo workflow updated
Why it matters: The commit message reads "Use latest version of Basic LLM Chain, to avoid fallback model issues". - 4January 6, 2026 — Last change to the compose file
Why it matters: The commit "Skip demo import if workflows already exist" changed `docker-compose.yml` so that a restart does not re-import the demo into an n8n instance that already has workflows. - 5July 23, 2026 — Latest commit on the default branch
Why it matters: The commit "Fixed broken link in the readme. No significant changes" touched only `README.md`.
What Can You Do With the Starter Kit?
The starter kit's value is in what the bundled services do together, not in new functionality of its own. Here is what each part actually does, per the official GitHub README.
- Local workflow automation — build and run automations in n8n's visual editor, using any of its 400+ integrations alongside dedicated AI/LangChain nodes for building agents
- Local API server — every bundled n8n instance exposes n8n's own REST API, so workflows can be triggered or managed programmatically from other local services, not only from the n8n UI
- Local LLM inference — Ollama runs open-weight models (Llama, Qwen, Gemma, and others) locally and is available to n8n workflows as a model provider with no external API key; the compose file pulls
llama3.2automatically on first start - Local vector storage for RAG — Qdrant stores and retrieves embeddings, and the kit pre-creates a "Local QdrantApi database" credential in n8n; the shipped demo workflow does not use it, so a RAG pipeline is something you build by adding Qdrant as the vector store in your own workflow
- Persistent workflow and credential storage — PostgreSQL backs n8n's own database, so workflows, credentials, and execution history persist across container restarts
- GPU acceleration profiles — separate Compose profiles (
cpu,gpu-nvidia,gpu-amd) let Ollama use available GPU hardware for faster local inference where supported - Shared file access — a mounted
/data/sharedfolder lets n8n read and write local files through nodes such as Read/Write Files from Disk, Local File Trigger and Execute Command, useful for document-ingestion workflows - Pre-loaded demo workflow — the kit imports one workflow ("Demo workflow": Chat Trigger, Basic LLM Chain, and an Ollama Chat Model set to
llama3.2:latest) on first start; open it and use the Chat button to talk to the local model
Usage Examples: Three Ways to Use the Starter Kit
The first example is the kit's own demo workflow. The other two are patterns you build yourself from the bundled parts, using nodes the README points to — the kit does not ship them as ready-made workflows.
Run the pre-loaded demo chat workflow
After docker compose --profile cpu up, open n8n at http://localhost:5678 and complete the one-time n8n setup. Open the included workflow, click Chat at the bottom of the canvas and send a message. On the first run, wait until Ollama finishes downloading llama3.2; the Docker logs show the progress. No cloud API key is involved.
Build a document-ingestion pipeline using the shared folder
Drop files into the mounted /data/shared folder, then build an n8n workflow that watches that folder (for example with the Local File Trigger node), embeds new documents with a local Ollama embedding model and indexes them in Qdrant for later retrieval. You assemble this pipeline yourself; the kit supplies the services and credentials, not the workflow.
Extend the base workflow with n8n's 400+ integrations
Start from the pre-loaded demo chat workflow, then add nodes for services you already use (email, Slack, a CRM, or a custom webhook) so that a local-model-powered response can trigger an action elsewhere — using n8n's broader integration catalog on top of the kit's local AI foundation.
Install the Self-Hosted AI Starter Kit
The starter kit installs via git clone and Docker Compose — there is no separate downloadable app. The steps below are from the official GitHub README — always verify against that page directly, since exact commands can change between updates.
Step | Command / Link |
|---|---|
| 1. Clone the repository | git clone https://github.com/n8n-io/self-hosted-ai-starter-kit.git |
| 2. Enter the directory | cd self-hosted-ai-starter-kit |
| 3. Copy the environment file | cp .env.example .env (then change the placeholder passwords and secrets) |
| 4a. Start (CPU-only) | docker compose --profile cpu up |
| 4b. Start (Nvidia GPU) | docker compose --profile gpu-nvidia up |
| 4c. Start (AMD GPU, Linux) | docker compose --profile gpu-amd up |
| 4d. Start (Mac / Apple Silicon) | docker compose up (Ollama runs on the Mac, see note) |
| Source code | github.com/n8n-io/self-hosted-ai-starter-kit |
n8n becomes available at http://localhost:5678 once the stack is up. Hardware needs scale with which models you run in Ollama — a small model works on a modest CPU-only machine, while larger models benefit from the Nvidia or AMD GPU profiles; check a given model's size in Ollama's library before pulling it. On a Mac with Apple Silicon, Docker cannot use the GPU: the README offers CPU-only operation, or Ollama installed on the Mac itself with OLLAMA_HOST=host.docker.internal:11434 set in .env, the "Local Ollama service" credential in n8n pointed at http://host.docker.internal:11434/, and docker compose up run without a profile. The .env.example ships placeholder secrets (the database password is password) that the README tells you to update. Docker and Docker Compose are required; the kit itself has no OS-specific build.
Pricing and Licensing
The starter kit itself is free and open-source, licensed under Apache-2.0. Since it bundles other open-source projects, it is worth knowing each bundled component's own licensing and cost model too.
Self-hosted AI Starter Kit (this template)
- License:
- Apache-2.0, per the repository LICENSE file
- Cost:
- Free
n8n (bundled)
- License:
- Sustainable Use License (source-available, not OSI-approved) for self-hosted use; see the n8n review for full licensing detail
- Cost:
- Free for self-hosted use under n8n's license terms
Ollama (bundled)
- License:
- MIT
- Cost:
- Free
Qdrant (bundled)
- License:
- Apache-2.0
- Cost:
- Free
PostgreSQL (bundled)
- License:
- PostgreSQL License (permissive)
- Cost:
- Free
Running the full stack costs nothing in software licensing — only your own infrastructure (compute, storage, and optionally a GPU) is a real cost. If you extend the pre-loaded n8n workflow to call a paid cloud LLM API or a paid third-party integration, that provider's own pricing applies; the starter kit itself does not add any markup or subscription fee.
Self-Hosted AI Starter Kit vs. Plain n8n
The starter kit is not a replacement for n8n — it is n8n plus a pre-wired local AI stack around it. The clearest way to think about the difference is what you get out of the box versus what you would have to configure yourself.
What you install
- Self-Hosted AI Starter Kit:
- n8n + Ollama + Qdrant + PostgreSQL, pre-networked in one docker-compose.yml
- Plain n8n Self-Hosted:
- n8n only — see the n8n review for its own install steps
Local LLM inference
- Self-Hosted AI Starter Kit:
- Included by default via bundled Ollama
- Plain n8n Self-Hosted:
- Not included; you connect your own model provider (local or cloud) manually
Vector database for RAG
- Self-Hosted AI Starter Kit:
- Included by default via bundled Qdrant
- Plain n8n Self-Hosted:
- Not included; you would add and network a vector database yourself
Example workflow
- Self-Hosted AI Starter Kit:
- One demo chat workflow (Chat Trigger, Basic LLM Chain, Ollama) plus pre-made Ollama and Qdrant credentials
- Plain n8n Self-Hosted:
- Starts with a blank canvas; templates are available separately from n8n's template library
Best for
- Self-Hosted AI Starter Kit:
- Getting a fully local AI workflow environment running in minutes, without manually wiring services together
- Plain n8n Self-Hosted:
- Teams that already run their own model or vector infrastructure and only need the automation layer
If you already have Ollama or a vector database running elsewhere in your infrastructure, installing plain n8n and connecting to your existing services may be simpler than adopting the starter kit's bundled versions. If you are starting from zero, the starter kit removes the integration work.
Who Should Use the Starter Kit?
Whether the starter kit fits depends on whether you want a ready-made local AI workflow environment versus assembling the same components yourself.
Use the Self-Hosted AI Starter Kit if
- You want to try local, agentic AI workflows (RAG, model-powered automations) without manually installing and networking a workflow tool, a model runner, and a vector database separately
- You are already interested in n8n and want an AI-focused starting point rather than a blank workflow canvas
- You want a working local chat-with-Ollama workflow to learn from and modify, rather than building one from a blank slate
- You need every piece of the stack to run on your own infrastructure with no cloud API key required to get started
- You have a GPU (Nvidia or AMD on Linux) and want a pre-configured Compose profile that uses it for local inference
Wait, or look elsewhere, if
- You already run n8n, Ollama, or a vector database independently and don't want the starter kit's bundled versions duplicating or conflicting with your existing setup
- You need a dedicated RAG framework with deeper document-processing features than what a workflow-automation tool provides — see the RAGFlow review or Haystack review for purpose-built alternatives
- You want n8n's self-hosted licensing terms (a source-available Sustainable Use License, not OSI-approved) to be a plain permissive open-source license for every bundled component — read n8n's own license directly if that distinction matters for your organization
- You need enterprise support or a production-grade deployment — the support channel is the n8n Forum, and n8n's documentation calls the kit "for testing only"
Starter Kit vs. Other Local AI Workflow Tools
The Self-hosted AI Starter Kit sits in the workflow-automation and RAG-framework space alongside several other self-hostable options. Here is how it compares to nearby tools — see the Local LLM Software Directory for the full catalog, and the dedicated Starter Kit vs. plain n8n comparison above for the closest head-to-head.
- n8n — the workflow-automation platform this kit is built on top of; see the dedicated comparison section above and the full n8n review for its own standalone install and licensing detail.
- LangChain-Chatchat — a self-hosted RAG and knowledge-base application with its own built-in UI, rather than a general-purpose workflow-automation canvas; see the LangChain-Chatchat review if a dedicated document-chat app fits your need better than a workflow builder.
- RAGFlow — a deep document-understanding RAG engine focused on parsing complex documents (tables, layouts) more thoroughly than a general workflow tool; see the RAGFlow review.
- Haystack — a Python-first RAG framework for developers who want to write pipeline code directly rather than use a visual workflow builder; see the Haystack review.
Common Mistakes When Evaluating the Starter Kit
Most confusion about the Self-hosted AI Starter Kit comes from treating it as a standalone application, misunderstanding what it actually bundles, or picking the wrong Compose profile.
Mistake 1: Treating it as a standalone app separate from n8n
The starter kit is not its own product — it is a Docker Compose template that installs and networks n8n alongside Ollama, Qdrant, and PostgreSQL. Any review of n8n itself (features, licensing, release cadence) applies to the n8n instance running inside this kit; see the n8n review for that detail.
Mistake 2: Assuming it includes a cloud LLM provider
The starter kit's default, bundled model provider is Ollama, running models entirely locally. It does not include or require any cloud LLM API key — if you want to add a cloud provider, you configure that separately inside an n8n workflow.
Mistake 3: Running the default cpu profile on a machine with a capable GPU
The Compose file ships three profiles — cpu, gpu-nvidia, and gpu-amd — and defaults to none unless you specify one on the command line. Running docker compose --profile cpu up on a machine with an Nvidia or AMD GPU wastes available acceleration; use the matching GPU profile instead. On a Mac with Apple Silicon no profile gives Docker access to the GPU; the README suggests running Ollama on the Mac itself.
Mistake 4: Expecting the starter kit to have its own independent version-release cadence
Because this is a configuration template rather than an application with its own codebase, it publishes no releases or tags. Do not expect a pinned stack: n8nio/n8n, ollama/ollama and qdrant/qdrant are pulled on latest, so versions change with the day you pull; only PostgreSQL is pinned (postgres:16-alpine).
Mistake 5: Treating the demo setup as production-ready
n8n's documentation labels the kit "for testing only", and the README says it is not fully optimized for production environments. The shipped .env.example uses placeholder secrets, and the compose file publishes ports 5678, 11434 and 6333 on the host; change the secrets and review the exposed ports before making the stack reachable beyond your own machine.
Frequently Asked Questions
What is the n8n Self-hosted AI Starter Kit?
It is an open-source Docker Compose template, curated by n8n and published at github.com/n8n-io/self-hosted-ai-starter-kit, that bundles n8n, Ollama, Qdrant, and PostgreSQL into one command for building local AI workflows with no cloud dependency required. Searches often call it the "n8n AI starter kit".
Is the Self-hosted AI Starter Kit free?
Yes. The template itself is licensed under Apache-2.0, and it costs nothing to run. Its bundled components (n8n, Ollama, Qdrant, PostgreSQL) are each free for self-hosted use under their own respective licenses — see the pricing table above.
What does the Self-hosted AI Starter Kit actually bundle?
Four separately maintained open-source projects: n8n (workflow automation, 400+ integrations), Ollama (local LLM inference), Qdrant (vector database), and PostgreSQL (relational database backing n8n's own data). It is a docker-compose configuration, not new software.
Does the Self-hosted AI Starter Kit require a cloud account or API key?
No. Every bundled service runs locally by default, and the demo workflow uses only the bundled Ollama model (llama3.2). You can add cloud provider API keys inside an n8n workflow later if you choose to.
How do I install the Self-hosted AI Starter Kit?
Clone the repository (git clone https://github.com/n8n-io/self-hosted-ai-starter-kit.git), copy the environment file (cp .env.example .env), then run docker compose --profile cpu up (or gpu-nvidia/gpu-amd for GPU acceleration). See the install table above for the full sequence.
Does it support GPU acceleration?
Yes, via separate Docker Compose profiles: gpu-nvidia for Nvidia GPUs and gpu-amd for AMD GPUs on Linux, alongside the default cpu profile for CPU-only inference.
Is this the same thing as n8n?
No. n8n is one of the four bundled components. The starter kit is a pre-configured environment that installs n8n alongside Ollama, Qdrant, and PostgreSQL — see the Self-Hosted AI Starter Kit vs. Plain n8n comparison above for the exact differences.
What license does n8n itself use inside this kit?
n8n is source-available under its own Sustainable Use License, which is not OSI-approved, for self-hosted use — distinct from the Apache-2.0 license covering this starter-kit template itself. See the n8n review for full licensing detail.
Is the Self-hosted AI Starter Kit ready for production?
No, according to n8n: its documentation labels the kit "for testing only", and the README says it is not fully optimized for production. Treat it as a proof-of-concept base: change the placeholder secrets in .env, review the exposed ports and pin your image versions before any production use.
Who maintains the Self-hosted AI Starter Kit?
It is curated and maintained by n8n, the company behind the n8n workflow automation platform, as a companion resource showcasing local AI workflows built with n8n's AI and LangChain nodes.
Is the Self-hosted AI Starter Kit still maintained?
The latest commit on the default branch dates from July 23, 2026 and only fixed a README link; the compose file last changed on January 6, 2026. The repository has no releases or tags, and n8n's documentation does not mention a deprecation. Check the repository before relying on it.
Does the Self-hosted AI Starter Kit run on a Mac with Apple Silicon?
Yes, but Docker cannot expose the Mac GPU. The README offers two options: run the whole kit on CPU (docker compose --profile cpu up), or install Ollama on the Mac, set OLLAMA_HOST=host.docker.internal:11434 in .env, point the "Local Ollama service" credential in n8n at http://host.docker.internal:11434/ and run docker compose up.