n8n-automations
N8N Automations
Businesses run on repetitive chores: scraping leads, chasing enquiries, sending invoices. n8n lets you build a chore as a flowchart of connected boxes that then runs itself, no code required. This is 59 such automations built for Digitalise Agency, each documented honestly, faults included: one supposedly eight-stage email campaign had only its first stage actually wired up. Real, still-working account keys found in the files were stripped.
Each workflow was pulled from the agency's live n8n instance or its on-disk exports, checked for real secrets and client-identifying data, scrubbed, and given its own README covering what it does, its architecture, and honest limitations found by tracing the actual node graph. A few names and course/tutorial-creator handles are withheld or corrected where the live workflow name was misleading or attributed to a third party.
Overview
A curated collection of n8n visual-workflow automations built for Digitalise Agency: lead generation and scraping, AI content and video pipelines, SEO and outreach systems, CRM and booking utilities. n8n builds automations as node graphs instead of text, but the engineering is the same, chaining triggers, API calls, LLM steps, and data transforms into a working system.
Digitalise Agency runs a live n8n instance with well over a hundred workflows built up over time: some production systems serving real clients, some demos, some abandoned tests. This project is the curation pass over that instance: survey every workflow, decide what is genuinely worth showing, pull its real definition, strip anything sensitive, and document it to the same standard as the rest of this portfolio, including the parts that do not work as advertised.
Six of the 59, as node graphs
n8n builds automations visually, so a code snippet does not really show the work. These are not screenshots: each is rendered directly from that automation's real exported workflow.json, using its exact node names, types, canvas positions, and connections, colored by node role (trigger, AI/LLM, logic, action). Ranging from a 6-node utility to a 190-node content pipeline.
Key features
- Lead generation and outreach: Google Maps business scrapers, a WhatsApp AI sales lead system with a Supabase RAG knowledge base, Instagram and LinkedIn lead-enrichment pipelines, a multi-stage cold-email sequence system.
- AI content pipelines: a faceless short-form video generator, an AI avatar video pipeline chaining ElevenLabs, HeyGen, an FFmpeg API, and Submagic captioning, a 190-node social media content generator, and several SEO blog-post generators.
- Agent systems: a Telegram-triggered personal assistant that routes to four sub-agents, a RAG ingestion pipeline into a Supabase vector store, an Upwork proposal-generation pipeline with a diagram-generating sub-workflow.
- Business operations: a Cal.com booking-and-reminder CRM pipeline, a CRM-to-Stripe invoicing workflow, booking-availability microservices across Calendly and Cal.com.
- Every workflow ships as an importable JSON file plus a README, a CREDENTIALS.md listing what external accounts it needs, and a PolyForm Strict LICENSE.
Verification
Every one of the 59 folders was checked for the required README, LICENSE, and CREDENTIALS.md files, every workflow JSON was validated with a JSON parser, and the whole collection was grepped for known API-key patterns, live webhook secrets, and personal email addresses. The pass caught real, unrotated credentials in several original source files (an Instantly.ai key, a Perplexity key, two Apify tokens, a Blotato key, a Supadata key, a ScreenshotOne key, a Fathom webhook secret), scrubbed in every copy here and flagged for the agency to rotate at the source. It also surfaced real defects in the underlying automations rather than hiding them: disconnected branches, a video pipeline that never writes its result back to its own tracking sheet, an "8-stage" email sequence where only the first stage is actually wired up live.
Tech stack
A challenge worth noting
n8n's workflow export strips credential bindings but not the hardcoded values some nodes had accumulated over time: real API keys sitting in HTTP Request node parameters, real Google Sheet IDs cached in node configuration, and once, a personal email address baked into pinned test data. There is no single grep pattern that catches all of that; it took reading each workflow's actual node graph rather than trusting an automated scan alone, since a URL query parameter or a credential-store `name` field can carry a real secret in a shape no generic pattern anticipates.