n8n 2 AI Agent Stack (Ollama + Qdrant)

A queue-mode n8n 2 stack with a local Ollama model and a Qdrant vector database on one private Railway network. It is for people who want AI agent workflows without sending data to external LLM APIs. Includes a webhook processor, workers, task runners and a demo chat workflow.

What you get

  • n8n 2.42.6 split into main, webhook processor, worker and isolated task runners for Code nodes
  • Postgres for workflows, credentials and executions, and Redis for the execution queue
  • Ollama pulling llama3.2:3b and nomic-embed-text on first boot (about 2.3 GB download)
  • Qdrant vector store protected by an API key
  • Ollama and Qdrant credentials plus a demo chat workflow seeded automatically
  • An nginx gateway as the only public service, routing production webhooks to the webhook processor

Good for

  • AI agents and chatbots on a local open model with tool calling
  • Retrieval-augmented workflows that store and search embeddings in Qdrant
  • High-volume webhook and API automations run by queue workers
  • Scheduled data syncs between SaaS tools, databases and internal APIs

Services in this template

  • Qdrant · image qdrant/qdrant:v1.19.2
  • n8n Webhook · image n8nio/n8n:2.42.6
  • n8n Task Runners · image n8nio/runners:2.42.6
  • Ollama · from baranberkay96/n8n-ai-stack-railway
  • Gateway · from baranberkay96/n8n-ai-stack-railway
  • n8n Main · image n8nio/n8n:2.42.6
  • n8n Bootstrap · from baranberkay96/n8n-ai-stack-railway
  • Postgres · image ghcr.io/railwayapp-templates/postgres-ssl:18
  • n8n Worker · image n8nio/n8n:2.42.6
  • Redis · image redis:8.2

Questions people ask

How do I get started after deploying?

Wait until n8n Main is healthy and the Ollama logs show both models pulled. Open the Gateway URL and create the owner account right away, then open the demo workflow Local AI chat (Ollama) and click Open chat. The Ollama and Qdrant credentials are already configured.

Can I use bigger models?

Set OLLAMA_PULL_MODELS on the Ollama service and redeploy. Railway has no GPUs, so keep local models small or use a hosted model API in the same workflows. The source notes that 7 to 8B models need about 6 to 8 GB RAM and are slow on CPU.

What drives the cost?

Ollama is the main factor, since it needs more than 1 GB of RAM and holds the model in memory while in use. The n8n services, Qdrant, Postgres and Redis are comparatively small. The source recommends the Hobby plan as the minimum.

How do I upgrade n8n?

Back up Postgres, then set the same new tag on n8n Main, Webhook, Worker, Task Runners and the Bootstrap Dockerfile. Runner and n8n versions must match. Ollama and Qdrant are pinned separately.

What licence applies?

n8n is fair-code under the Sustainable Use License: free to self-host for your own business, not to resell as a hosted service. Enterprise features need an n8n licence. The pulled llama3.2:3b model is under the Llama 3.2 Community License.

Full template README

Deploy and Host n8n with Railway

n8n is a fair-code workflow automation platform with a visual editor, 400+ integrations and native AI agent nodes. This community template runs n8n 2 in queue mode with a webhook processor, workers, isolated task runners, a local Ollama LLM and a Qdrant vector database, all on one private Railway network.

About Hosting n8n

A single n8n container works for experiments, but production automations need more: Postgres instead of SQLite, Redis-backed queue mode so workers execute runs while the editor stays responsive, a webhook processor for incoming traffic, and task runners that execute Code nodes outside the n8n process. AI agent workflows add a model server and a vector store for memory and retrieval. This template wires all of these with generated secrets, readiness checks and persistent volumes, pulls a small open model on first boot, and seeds ready-made Ollama and Qdrant credentials plus a demo chat workflow, so you can build private AI agents without sending data to external LLM APIs.

Common Use Cases

  • AI agents and chatbots that run on a local open model (Ollama) with tool calling
  • Retrieval-augmented workflows that store and search embeddings in Qdrant
  • High-volume webhook and API automations processed by queue workers
  • Scheduled data syncs between SaaS tools, databases and internal APIs
  • Exposing workflows as MCP servers or chat endpoints for other AI tools

Dependencies for n8n Hosting

  • n8n main, webhook processor and worker (n8nio/n8n)
  • n8n task runners (n8nio/runners) for Code nodes
  • PostgreSQL (workflows, credentials, executions, binary data)
  • Redis (BullMQ execution queue)
  • Ollama (local LLM and embeddings) and Qdrant (vector store)
  • nginx gateway (routes production webhooks to the webhook processor)

Deployment Dependencies

Implementation Details

Service Image Role Public
Gateway nginx 1.30 (repo build) Routes /webhook, /webhook-waiting, /form, /form-waiting, /mcp to the webhook processor, everything else to main yes
n8n Main n8nio/n8n:2.42.6 Editor, REST API, triggers, test executions (offloaded to workers) no
n8n Webhook n8nio/n8n:2.42.6 (n8n webhook) Production webhooks, forms, chat and MCP triggers no
n8n Worker n8nio/n8n:2.42.6 (n8n worker --concurrency=10) Executes workflows from the Redis queue no
n8n Task Runners n8nio/runners:2.42.6 Runs JavaScript and Python Code nodes for the worker no
Ollama ollama/ollama:0.40.3 (repo build) Local LLM + embeddings; pulls llama3.2:3b and nomic-embed-text:v1.5 on first boot no
Qdrant qdrant/qdrant:v1.19.2 Vector store (API key protected) no
n8n Bootstrap n8nio/n8n:2.42.6 (repo build) One-shot: seeds Ollama/Qdrant credentials and a demo workflow, then exits no
Postgres / Redis Railway base images Database / queue no

First login

  1. Wait until n8n Main is healthy and the Ollama logs show both models pulled (first boot downloads about 2.3 GB).
  2. Open the Gateway URL and create the owner account (n8n's first-run setup; do this right after deploying).
  3. Open the workflow Local AI chat (Ollama) - demo and click Open chat. Credentials Ollama (Railway) and Qdrant (Railway) are already configured for the private services.

Scaling workers: raise --concurrency in the worker start command first. For more workers, duplicate the n8n Worker + n8n Task Runners pair and point the new runner's N8N_RUNNERS_TASK_BROKER_URI at the new worker; each worker needs its own runner. The webhook processor can run multiple replicas.

Models: change OLLAMA_PULL_MODELS on the Ollama service (comma-separated) and redeploy to pull more models. Railway has no GPUs, so keep local models small or use a hosted model API in the same workflows.

Custom domain: add it to the Gateway, then update WEBHOOK_URL and N8N_EDITOR_BASE_URL on n8n Main (other n8n services reference these values).

Pinning and upgrading: n8n images are pinned to 2.42.6. To upgrade, back up Postgres and set the same new tag on n8n Main, n8n Webhook, n8n Worker, n8n Task Runners and the Bootstrap Dockerfile (runner and n8n versions must match). Ollama and Qdrant are pinned separately.

License note: n8n is fair-code under the Sustainable Use License: free to self-host for your own business, not to resell as a hosted service. Enterprise features (.ee code) need an n8n license.

Why Deploy n8n on Railway?

Railway runs the full n8n queue-mode stack, a local LLM and a vector database in one project on a private network with managed HTTPS and persistent volumes. You get a production-shaped AI automation platform without managing servers, can add workers or memory from the dashboard, and roll back any deploy in one click. This is a community template and is not affiliated with n8n GmbH, Ollama or Qdrant.

Guides for this template

View n8n 2 AI Agent Stack (Ollama + Qdrant) on railway.com

Similar templates

TemplateCategoryDeploymentsDeploys
N8N (w/ workers)
A powerful workflow automation tool for technical people
Automation114,737
n8n
n8n [Nov '26] (Workflows/Integrations/API Automation/AI Coding) Self Host
Automation17,560
Typebot
Build advanced chatbots visually
Automation8,968
n8n
n8n workflow automation on PostgreSQL, webhooks and secrets preconfigured
Automation5,850
n8n with workers
A complete setup for n8n with workers
Automation4,040
N8N (w/ webhook processors)
A powerful workflow automation tool for technical people
Automation3,386