Visual AI Agent Builder (Langflow + Ollama)

A low-code visual framework for building AI applications. Langflow gives you a drag-and-drop canvas to construct multi-agent systems, RAG pipelines, and chatbot workflows - then export them as production-ready APIs. Connect it to Ollama for local model inference, and you can prototype and deploy complex AI pipelines entirely on your own hardware.

Langflow's real-time visual feedback shows data flowing between nodes as you build, making it one of the most intuitive tools for designing AI agent architectures.

What you get

• Drag-and-drop AI canvas - visually connect LLMs, vector stores, agents, and tools
• Multi-agent orchestration - build systems with multiple specialized agents that collaborate
• RAG pipeline builder - document ingestion, chunking, embedding, retrieval, all visual
• Real-time testing - run individual nodes or full flows and see data flow live
• Export as API - every flow gets a REST API endpoint you can call from your app
• MCP support - connect Model Context Protocol servers as tools for your agents
• $0/mo - all local, no API keys needed

Architecture

Recommended GPU: RTX 3060 12GB for 14B models, or RTX 4070 Super for Qwen3 30B A3B (MoE, fast).

Prerequisites

• A GPU with ≥12 GB VRAM for local LLM inference (CPU-only mode works for flow prototyping)
• Python 3.10+ or Docker
• 4 GB RAM minimum
• ~5 GB free disk

Setup

Option A: Docker (Recommended)

Save this as docker-compose.yml:

services:
  ollama:
    image: ollama/ollama:latest
    container_name: ollama
    volumes:
      - ollama:/root/.ollama
    ports:
      - "11434:11434"
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    restart: unless-stopped
 
  langflow:
    image: langflowai/langflow:latest
    container_name: langflow
    depends_on:
      - ollama
    ports:
      - "7860:7860"
    volumes:
      - langflow_data:/app/langflow
    environment:
      - LANGFLOW_AUTO_LOGIN=true
    restart: unless-stopped
 
volumes:
  ollama:
  langflow_data:

Launch it:

docker compose up -d
docker exec ollama ollama pull qwen3:14b

Open http://localhost:7860 to access Langflow.

Option B: pip Install

pip install langflow
langflow run

Then install and run Ollama separately:

# Install Ollama from https://ollama.com
ollama pull qwen3:14b

Open http://localhost:7860 for Langflow.

Connect Langflow to Ollama

In Langflow, add these components to your flow:

1. Ollama Chat Model - set Base URL to http://ollama:11434 (Docker) or http://localhost:11434 (pip)
2. Select model: qwen3:14b
3. Connect it to a Prompt node and Chat Output for a basic chatbot

Use it

Build a Chatbot with RAG

1. Drag in: File → Ollama Embeddings → Vector Store (Chroma) → Ollama Chat Model → Chat Output
2. Upload a PDF to the File component
3. Ask questions about the document - answers come from your local knowledge base
4. Export as API endpoint for your frontend

Multi-Agent Research System

1. Create a Agent node with a Web Search Tool connected to Ollama
2. Add a second Agent node for summarization
3. Use a Chat Input to receive the query, route through both agents
4. The search agent gathers info, the summary agent condenses it

Document Processing Pipeline

1. Combine File Loader → Splitter → Ollama Embeddings → Vector Store
2. Add Ollama Chat Model with a custom prompt template
3. Build a question-answering system over your documents

Cost vs cloud

Troubleshooting

• Langflow shows "Model not found" → Make sure the model name matches exactly what you pulled in Ollama. Run docker exec ollama ollama list to see available models.
• Slow responses → For faster inference, try Qwen3 30B A3B (MoE - only 3B active parameters per token) or a smaller model like Llama 3.1 8B.
• Can't upload files → Langflow stores uploads in its container volume. For larger files, mount a host directory.
• Docker networking → If Langflow can't reach Ollama, use http://host.docker.internal:11434 (Docker Desktop) or the container name ollama:11434 (same compose file).

Swap components

• Use Qdrant instead of Chroma → Add Qdrant as a Docker service and use the Qdrant vector store component.
• Hybrid cloud/local → Add an OpenAI or Anthropic component alongside Ollama for different tasks.
• Try Flowise → Flowise is a similar visual builder with a different node design philosophy.
• Production API → Set LANGFLOW_AUTO_LOGIN=false and configure authentication via environment variables.