---
title: n8n Self-Hosted AI Starter Kit
emoji: 🚀
colorFrom: green
colorTo: blue
sdk: docker
app_port: 5678
pinned: true
# Add hardware requests if needed, e.g. for Ollama with larger models
# hardware: cpu-upgrade # or gpu-small / gpu-a10g-small etc.
# For GPU, Ollama should pick it up if drivers are compatible.
# This Dockerfile is built for CPU by default for wider compatibility.
persistent_storage:
  # Request persistent storage. Adjust size as needed for models, db, etc.
  # n8n data, postgres db, qdrant vectors, ollama models
  # Example: 10Gi for models and data
  data:
    type: "disk"
    size: 10 # GiB
    mount_path: "/data" # This is where it will be mounted in the container
---

# n8n Self-Hosted AI Starter Kit on Hugging Face Spaces

This Space deploys the [n8n Self-Hosted AI Starter Kit](https://github.com/n8n-io/self-hosted-ai-starter-kit) in a single container environment. It includes:

*   **n8n**: Low-code automation platform.
*   **Ollama**: Run local LLMs.
*   **Qdrant**: Vector database.
*   **PostgreSQL**: Database for n8n.

## Setup

1.  **Secrets**:
    Before this Space can run correctly, you **MUST** configure the following secrets in your Space settings:
    *   `N8N_ENCRYPTION_KEY`: A long, random string for encrypting sensitive n8n data.
        *   Generate one with: `openssl rand -hex 32`
    *   `DB_POSTGRESDB_PASSWORD`: Password for the PostgreSQL `n8n` user and database.
        *   Choose a strong password.

2.  **Wait for Build**: The first time, the Docker image will build, and services will start. This can take a few minutes. The `start.sh` script will also pull a default Ollama model (`llama3.1:8b` or similar small model), which can take time.

3.  **Access n8n**: Once the Space is running, click the "Open app" button or use the direct URL to access n8n on port 5678.

## How to Use

1.  **n8n First-Time Setup**: When you first access n8n, you'll be prompted to create an owner account.
2.  **Local Services**:
    *   **Ollama**: Running on `http://localhost:11434` (internally within the Space). The default model (`llama3.1:8b` or similar) should be available. You can use the "Ollama" node in n8n and set its "Base URL" to `http://localhost:11434`.
    *   **Qdrant**: Running on `http://localhost:6333` (internally). Use the "Qdrant" node in n8n and set its "URL" to `http://localhost:6333`.
    *   **PostgreSQL**: n8n is already configured to use the internal PostgreSQL instance.
3.  **Persistent Storage**:
    *   n8n workflows and credentials are stored in `/data/.n8n`.
    *   PostgreSQL data is stored in `/data/postgres`.
    *   Qdrant vector data is stored in `/data/qdrant_storage`.
    *   Ollama models are stored in `/data/.ollama`.
    This data will persist across Space restarts if you've configured persistent storage for `/data`.

4.  **Accessing Local Files in n8n**:
    The original starter kit mentions a `/data/shared` directory for n8n to access local files. This directory is available in this Space. When using n8n nodes like "Read/Write Files from Disk", use paths like `/data/shared/yourfile.txt`.

## Important Notes & Limitations

*   **Single Container**: All services (n8n, Postgres, Qdrant, Ollama) run in a single Docker container. This is less robust than a multi-container `docker-compose` setup but necessary for this Hugging Face Space deployment model.
*   **Resource Usage**: Running all these services, especially Ollama with LLMs, can be resource-intensive. Ensure your Space has adequate CPU, RAM, and storage. Consider upgrading hardware if performance is slow.
*   **Ollama Model Management**:
    *   A default model is pulled on startup.
    *   To pull other models, you would typically exec into the container (possible via `hf cli` or other tools, but not directly via the Space UI) and run `ollama pull <modelname>`. Alternatively, some n8n Ollama nodes might trigger a download if the model isn't present (behavior may vary).
*   **Initial Startup Time**: The first startup might be slow due to database initialization and model download. Check the logs for progress.
*   **WEBHOOK_URL / N8N_EDITOR_BASE_URL**: These are automatically configured by the `start.sh` script to use the Hugging Face Space URL.

## OAuth Configuration (ClickUp and Other OAuth Providers)

This n8n instance is configured to use `n8n-clickup-oauth-proxy.vercel.app` as the base URL for OAuth callbacks and webhooks.

**For ClickUp OAuth setup in n8n:**
- **Redirect URL**: `https://n8n-clickup-oauth-proxy.vercel.app/rest/oauth2-credential/callback`
- Ensure your proxy at `n8n-clickup-oauth-proxy.vercel.app` forwards requests to the Space's actual n8n endpoint: `https://leon4gr45-n8n.hf.space/rest/oauth2-credential/callback`

**Alternative webhook URL** (if needed):
- `https://n8n-clickup-oauth-proxy.vercel.app/rest/workflow-callback-url`

## Troubleshooting

*   Check the Space logs for any errors during startup or operation.
*   Ensure secrets are correctly set.
*   If n8n is inaccessible, one of the backend services might have failed to start.