n8n Workflow Manager
This article profiles n8n as deterministic workflow infrastructure: a source-available, node-based automation platform enabling sovereign orchestration of AI agents without vendor lock-in. It catalogs 10 agent archetypes—email support, RAG knowledge-base, lead enrichment, research, voice, code-review, document-processing—each mixing LLM nodes (OpenAI, Claude, Gemini, Ollama) with 500+ integrations via visual canvas. Framed within your Unification Project, n8n treats process automation as lawful informational architecture: value flows from individual verification and reproducible protocols, not imposed doctrine. Self-hosting ensures user control; modular nodes enable testable, scalable reasoning. The platform shifts AI from recurring API expense to capital investment in deterministic, user-sovereign workflow—where intelligence is public good, not proprietary service.
Rui Manuel de Almeida Pinheiro
Mainframe Analyst. Prompt Engineering. Content Engineering. Framework Design.
December 26, 2025
Let’s jump to easy things people like me understand and can use.
n8n (pronounced “n-eight-n”) is a source-available workflow automation platform that lets you connect apps and services, move data, and orchestrate repetitive tasks—without writing code—via a visual node-based editor. You can self-host it for free or use its managed cloud service.
Below are the most popular AI-agent patterns that people build with n8n today.
(They are not a fixed “product line”; instead you mix-and-match the built-in Agent node, LLM nodes and 500+ integrations to create any of these archetypes.)
1. Email-support agent
Trigger: new Gmail ticket
Brain: GPT-4 / Claude / Gemini
Actions: classify, draft reply, ask human for approval, send, log to Sheets
Template pack: n8n.io
2. Company knowledge-base (RAG) agent
Ingest: Google Drive, Notion, Confluence → vector DB (Pinecone, Qdrant, Weaviate)
Chat interface: Slack, Teams or Telegram
Brain: any LLM + Retrieval plugin
Example: HR or Legal Q&A bot
3. Lead-enrichment & personalised outreach agent
Trigger: new row in Google Sheet or CRM
Steps: scrape website → summarise with LLM → craft personalised email → send via Gmail/Outlook → update HubSpot / Pipedrive
Pack:
4. Social-content agent
Input: topic or RSS feed
Generate: tweet / LinkedIn post / Instagram caption with GPT-4o or Gemini
Schedule: Buffer, Hootsuite, or native APIs
Template: n8n.io
5. Multi-source research agent
Firecrawl + SerpAPI → pull pages → LLM summary → push digest to Slack / email
Example: weekly competitor or arXiv trend alerter
6. Voice / phone agent (VAPI)
Telephony trigger → Speech-to-Text → LLM → Text-to-Speech
Use-case: restaurant booking, help-desk IVR
Infrastructure node exists
7. Code-review & PR-risk agent
Trigger: GitHub / GitLab pull request
Claude or GPT analyses diff → posts risk score + recommendations → logs to Google Sheet
Template:
8. Document-processing agent
Inbox receives PDF invoice → OCR (Tesseract or AWS Textract) → LLM extracts fields → Google Sheets → Slack approval → QuickBooks posting
Example: n8n.io
9. Crypto trading / sentiment agent
Stream news → sentiment with GPT-4o → conditionally place Binance order via API → alert Telegram
Template: n8n.io
10. Autonomous “multi-step” task agent
Uses n8n’s Loop, Wait, Evaluate nodes so the LLM can create sub-tasks, execute them, check results and iterate until goal is met (BabyAGI-style)
Build guide:
Quick-start kits you can import
“Starter kit” – Weather & web-scraping demo agent
20 B2B / B2C templates (DM personaliser, email classifier, etc.) – free pack
Here are the LLM nodes you can drop straight onto an n8n canvas today (v 1.73+):
Chat & Language Model Providers
Provider: OpenAI Chat Model
Models: GPT-4o, GPT-4o-mini, GPT-4-turbo, GPT-3.5-turbo, DALL·E, TTS…
Notes: Also works for any OpenAI-compatible endpoint by changing the base-url
Provider: Azure OpenAI Chat Model
Models: Same models as above, routed through your Azure subscription
Provider: Anthropic Chat Model
Models: Claude 3.5 Sonnet, Claude 3 Haiku, Claude 2
Provider: Google Gemini Chat Model
Models: Gemini 1.5 Pro, 1.5 Flash, Pro
Provider: Google PaLM Chat Model
Models: PaLM 2
Notes: Legacy, still present
Provider: Google Vertex AI Chat Model
Models: Gemini & PaLM 2 via Vertex
Provider: Cohere Chat Model
Models: Command, Command-Light, Command-R, Command-R+
Provider: DeepSeek Chat Model
Models: DeepSeek-Chat, DeepSeek-Coder
Provider: Mistral Cloud Chat Model
Models: Mistral 7B, Mixtral 8x7B, Mistral-Large
Provider: Ollama Model
Models: Any model you pull locally (Llama 3, Phi-3, Mistral, Gemma …)
Notes: Easiest on-prem option
Provider: AWS Bedrock Chat Model
Models: Claude, Llama 2, Titan, Jurassic, Command…
Provider: Hugging Face Inference
Models: Thousands of models hosted on HF (fill-mask, text-gen, etc.)
Provider: OpenRouter (via OpenAI node)
Models: Gives you 100+ OSS & closed models with one credential
Notes: Just set base-url to https://openrouter.ai/api/v1
Provider: LLM Gateway (via OpenAI node)
Models: GPT-5, Claude-4, Gemini Ultra, etc.
Notes: Set base-url to https://api.llmgateway.io/v1
Utility & embedding nodes
Embeddings nodes for every provider above (generate vectors for RAG)
Basic LLM Chain – quick prompt → completion wrapper
AI Agent – LangChain-powered agent that can pick any of the chat models and add tools, memory, retrieval, etc.
Need a provider that isn’t listed?
Drop in the HTTP Request node and call its REST endpoint directly—n8n will still treat the response like any other node output.
That’s the complete current “LLM node” lineup.