pkuppens/agent-visual-dev

Agent Visual Development

Patterns for building AI agents with visual/GUI tools instead of (or alongside) code. Covers n8n, Flowise, LangFlow, and when visual development is the right choice.

When to use

• Evaluating visual agent builders for rapid prototyping
• Non-developers need to create or modify agent workflows
• Building agent pipelines where visual flow representation aids understanding
• Comparing visual tools vs code-first frameworks
• Self-hosting visual tools on-premises

Tool comparison

| Tool | Focus | Self-hosted | LLM integrations | Agent capabilities | |------|-------|-------------|-------------------|--------------------| | n8n | General workflow automation + AI | Yes (Docker, npm) | OpenAI, Ollama, Anthropic, Azure | AI agent node, tool use, memory, sub-workflows | | Flowise | LangChain visual builder | Yes (Docker, npm) | All LangChain providers | Chains, agents, RAG, tools, chatflows | | LangFlow | LangChain/LangGraph visual builder | Yes (pip, Docker) | All LangChain providers | Flows, agents, RAG, custom components | | Dify | LLM app platform | Yes (Docker) | OpenAI, Ollama, Azure, HuggingFace | Agents, RAG, workflow orchestration | | Rivet | Visual AI pipeline editor | Desktop app | OpenAI, Anthropic | Graph-based prompt chains, evaluation |

When visual vs code

| Factor | Visual tools | Code-first | |--------|-------------|------------| | Prototyping speed | Fast — drag and drop | Slower for initial setup | | Complex logic | Limited — hard to express conditional branching | Full control | | Version control | Limited — JSON export, harder to diff | Native git workflow | | Testing | Manual via UI | Automated test suites | | Debugging | Visual trace, step-by-step | Standard debugging tools | | Team scaling | Good for small teams, non-devs | Better for engineering teams | | Production deployment | Possible but less mature | Standard CI/CD pipelines | | On-premises | All tools above support self-hosting | Full control |

Recommended approach

• Prototype with visual tools to validate the agent design
• Implement in code for production systems that need testing, CI/CD, and version control
• Hybrid — use n8n for workflow orchestration, call code-based agents as HTTP services

n8n for AI agents

n8n stands out as the most general-purpose option with strong AI capabilities:

• AI Agent node — built-in ReAct agent with tool use
• Memory nodes — conversation history (buffer, window, vector store-backed)
• Tool nodes — HTTP request, code execution, database queries, calculator
• RAG workflow — document loaders, text splitters, embeddings, vector stores
• Sub-workflows — compose complex agent pipelines from reusable pieces
• Self-hosted — Docker or npm install; runs entirely on-premises
• Credentials — manage API keys securely with built-in credential store

n8n Docker setup (on-premises)

services:
  n8n:
    image: n8nio/n8n:latest
    ports:
      - "5678:5678"
    environment:
      N8N_BASIC_AUTH_ACTIVE: "true"
      N8N_BASIC_AUTH_USER: admin
      N8N_BASIC_AUTH_PASSWORD: changeme
    volumes:
      - n8n_data:/home/node/.n8n

Flowise and LangFlow

Both are visual frontends for LangChain:

• Flowise — Node.js based, simpler UI, focuses on chatflows and RAG
• LangFlow — Python based, closer to LangChain/LangGraph API, supports custom Python components

Choose Flowise for quick RAG chatbots. Choose LangFlow when you need LangGraph integration or custom Python components.

Self-hosting checklist

• [ ] Deploy with Docker Compose (all tools above support it)
• [ ] Configure persistent storage for workflows and credentials
• [ ] Set up authentication (basic auth, OAuth, or reverse proxy)
• [ ] Connect to local LLM server (Ollama) — no cloud API needed
• [ ] Back up workflow definitions (export JSON, or mount config volume)
• [ ] Monitor resource usage (CPU, memory, GPU if applicable)

Integration with other skills

• agent-frameworks — visual tools complement code-first frameworks
• agent-on-premises — self-hosting visual tools locally
• agent-llm-providers — configuring LLM providers in visual tools
• deployment-build — Docker-based deployment of visual tools