Heimdall

[Feb '26] Integrate self-hosted LLM models into N8N workflows.

N8N with Ollama |  Build private, self-hosted AI automation workflows

worker

Time (ms) before timeout during db connection

The name of the PostgreSQL database that stores n8n’s workflow and execution data.

Host address of the PostgreSQL database connected to n8n.

Password for authenticating the PostgreSQL connection.

Port number where PostgreSQL is running — usually 5432.

Username used by n8n to authenticate with PostgreSQL.

Specifies the type of database used by n8n — here it’s postgresdb.

Enables private networking between services on Railway’s infrastructure for secure internal communication.

Determines how workflows are processed. queue enables distributed execution using Redis.

Secret key used to encrypt sensitive credentials stored in the n8n database.

Ensures configuration files have strict permissions for improved security in production environments.

Defines which network interfaces n8n binds to — :: allows both IPv4 and IPv6.

Enables “Runners,” allowing distributed workflow execution across multiple worker nodes.

Runs manual (UI-triggered) executions on worker nodes instead of the main instance to improve performance.

Points N8N to the internal Ollama API endpoint for model access.

Port on which the n8n service listens (typically 5678).

Enables dual IPv4 and IPv6 support for Redis connections.

Redis host address used by n8n for managing job queues in queue mode.

Password used to connect securely to Redis.

Username used for authenticating Redis connections (if required).

Public URL endpoint for receiving external webhook triggers through Railway’s public domain.

Ollama

You can specify the model to download at boot time (comma separated, if multiple). This can be updated from UI later as well. By default it downloads the deepsek-r1:1.5b model(from https://ollama.com/library). Please make sure you have enough memory to download the specified model.

Allows Ollama to listen on all interfaces

Comma-separated list of allowed origins for CORS.

main

Automatically deletes old workflow execution logs to save storage space.

Maximum number of workflow execution records to retain before pruning older ones.

Defines where binary files (images, PDFs, etc.) are stored — filesystem saves them to attached storage.

Base URL where the n8n editor (web UI) is accessible.

Internal port for n8n’s web interface and API (usually same as PORT).

Configures Node.js memory usage — This config (8192) allows n8n to use up to 8 GB of memory.

Postgres

The publicly accessible connection string for external database access.

The full connection string to connect to Postgres

The directory path inside the container where PostgreSQL stores its data files

Name of the primary database to connect to.

The hostname or internal private domain for connecting to PostgreSQL within Railway’s private network. Used by other connected services like n8n.

Password used for connecting to the database

The port PostgreSQL listens on (default: 5432). Often left empty here because Railway sets it automatically.

The name of the default database created during initialization

Auto-generated secure 32-character password for the Postgres instance

The username of the Postgres superuser account created when the container starts

Defines the grace period (in seconds) Railway waits before fully stopping a service during redeployments.

Defines how long SSL certificates should be valid if PostgreSQL is using SSL mode.

Redis

The private domain address of your Redis instance — used for internal service communication within Railway.

The port on which Redis is running — typically 6379.

The username used for authenticating connections to Redis.

Securely generated password (using Railway’s secret() helper) used to authenticate Redis connections.

The public URL for connecting to Redis externally.

The full connection URL for Redis, usually combining host, port, and authentication details.