Introduction

Cloud AI agent platforms are a scam.

Services charge you for every single “task” or “step” executed by an autonomous agent. You write a simple prompt to search the web and summarize data, and suddenly you are hitting a paywall because your AI looped 500 times in a day.

Now comes the real part: Running a fully containerized, offline-capable n8n Community Edition deployment featuring an autonomous AI Assistant sandbox and local web search.

This blog explains how to deploy an n8n Local AI Stack on your machine. It uses Docker to give you a secure DinD (Docker-in-Docker) environment for your AI to execute code autonomously, completely bypassing cloud subscription limits.

Local n8n Docker Setup

What This Setup Achieves

By containerizing this AI stack locally, we get:

  • Absolute zero cost. Free forever.
  • Unlimited workflow executions (N8N_ENFORCEMENT_EDITION=community).
  • Private, local metasearch engine (SearXNG) allowing AI agents to query the live internet without requiring paid APIs.
  • AI Code Sandbox: A Docker-in-Docker environment (sandbox-runner-1) that allows the n8n Assistant to safely write and execute code autonomously.
  • Easy workflow management: Automatically sync local JSON workflows into the n8n SQLite database on boot.
  • A portable offline bundle that can be run completely offline using cached Docker images.

Minimal architecture, infinite automation.

Prerequisites

  • Docker Engine and Docker Compose v2 installed
  • A free OpenRouter API key (for the AI Assistant model)
  • Another free Google AI studio API key (for the Google Search model)

Step 1: Clone the Repo & Configure Environment

Do not run massive, unreadable CLI commands from scratch. Let’s pull the pre-configured repository.

git clone https://github.com/backtovedas/n8n-local.git
cd n8n-local

Ensure your .env contains your API keys for the AI providers and the internal container secrets for mTLS.

# Internal Container Secrets (used for mTLS / container auth)
SANDBOX_API_KEYS=sandbox-secret-key-92837492
SANDBOX_API_RUNNER_REGISTRATION_TOKEN=runner-token-47291047
SANDBOX_API_RUNNER_API_KEY=runner-key-83746291

# Must match SANDBOX_API_KEYS so n8n can authenticate to sandbox-api
N8N_SANDBOX_SERVICE_API_KEY=sandbox-secret-key-92837492

# Secret key for the local SearXNG metasearch instance
SEARXNG_SECRET=searxng-secret-58294710

# AI Provider API Key (e.g., OpenRouter, OpenAI, or Anthropic)
N8N_INSTANCE_AI_MODEL_API_KEY=sk-or-v1-your-openrouter-key-here

Step 2: Start the Stack

To spin up the entire autonomous environment:

docker compose up -d

Open http://localhost:5678 in your browser. You will be prompted to set up a local owner account. This is strictly local; it does not connect to the cloud.

The Container Architecture

This stack is composed of several specialized containers working together:

  • sandbox-certs: Runs once on boot to generate mutual TLS certificates for secure container communication, then exits.
  • sandbox-api: The control plane that n8n communicates with to request code execution.
  • sandbox-runner-1: Runs in privileged mode. It spins up temporary, isolated sub-containers to execute AI-generated code, then destroys them.
  • n8n-import: An init container that automatically pushes local JSON files from ./workflows/ into n8n on startup.
  • searxng: The private, local metasearch engine configured to output JSON for the AI agents.

Managing Workflows (Syncing)

n8n does not natively read workflows directly from the filesystem; they live in an internal SQLite database. We use CLI commands to sync them.

1. Auto-Import on Boot (Local -> n8n)

When you run docker compose up -d, the n8n-import init container runs first. It scans the ./workflows/ directory and safely pushes any .json files into the n8n database before the main application boots.

2. Manual Export (n8n -> Local)

When you create or modify a workflow inside the n8n web UI, it is saved in the database, not your local folder. To back up your current workflows to the ./workflows/ directory, run:

docker container exec -it n8n n8n export:workflow --all --output=/workflows/

(Note: Exported JSON files contain node configurations but do not export secure credentials like API keys/passwords for security reasons).

Stopping the Stack

To shut down the entire environment safely without losing your persistent data:

docker compose down

Conclusion

Automating your digital life and running autonomous AI agents should not require a subscription model.

By pulling this localized Docker stack, you get an enterprise-grade AI automation environment running entirely on your own hardware, completely detached from corporate rate limits.

Credits (Important)

Before anything else:

  • The n8n Team – for keeping the core automation engine source-available and free to self-host.
  • Docker – for making application deployment a single text file instead of a dependency nightmare.
  • SearXNG – for a privacy-respecting, hackable metasearch engine.

Without these, we would all be paying $30 a month just to let an AI scrape a single webpage.

Personal Opinion

I refuse to pay per-execution fees for AI code that runs on my own hardware. This setup exists so I can spin up autonomous AI agents, let them write code in an isolated sandbox, query the web without API limits, and automate my workflows without counting how many “tasks” I have left this month. It is self-hosted, it runs offline, and it never asks for a credit card — and that is exactly how software should be.