Flowise

Flowise Agent Flow with Llama3.1 Chat
Flowise Agent Flow with Llama3.1 Chat

Flowise is a visual no-code AI workflow builder used to create LLM pipelines, RAG systems, chatbots and agent flows through a drag-and-drop UI.

It provides a browser-based interface that makes it easy to combine LLMs, vector databases, APIs, tools, and custom logic into complete applications. Flowise focuses on rapid prototyping and operational deployment rather than low-level model management. It is open source, modular, and supports both local and cloud environments.

Key Use Cases

Example Flowise Install in your GPU Server
Example Flowise Install in your GPU Server

Flowise can be used to build:

  • Chatbots & Assistants: Custom conversational agents for internal or external use.
  • RAG Pipelines: Document search and question answering powered by embeddings + vector DBs.
  • Autonomous Agents: Task-driven workflows triggered by external APIs or business logic.
  • Data Processing: Summarization, sentiment analysis, classification, extraction.
  • Business Automation: Lead qualification, customer support routing, content workflows.

Flowise also integrates with major vector databases (Qdrant, Milvus, Chroma, Pinecone), local LLM engines (Ollama), and cloud LLM providers (OpenAI, Anthropic, Mistral).

How to Configure local Ollama in Flowise

Example config of Ollama in Flowise
Example config of Ollama in Flowise

Install the OpenWebUI & Ollama to run your local Ollama service. This setup is optimized for a single GPU server and integrates directly with Flowise.

Configuration Notes

  • Node Model: select ChatOllama

  • Base URL:
    If Flowise runs in a separate container →
    http://host.docker.internal:11434
    (Use this instead of localhost, since localhost would refer to the container itself.)

  • Model Name:
    Ensure the model listed in Flowise exactly matches the name in Ollama/OpenWebUI.

  • Temperature:
    A value around 0.3 is recommended for factual tasks.
    Higher values (e.g., 0.8) increase creativity but also hallucination risk.

Notes on GPU Usage

Flowise itself does not require a GPU.
GPU usage depends on the models behind Ollama or other LLM backends.

Qdrant / Vector DB Integration

For RAG pipelines, Flowise includes Retriever Vector DB nodes: select database type (e.g., Qdrant), then configure:

  • URL: http://host.docker.internal:6333
  • Collection name
  • Embedding model source

Typical Flowise Troubleshooting

1. Flowise UI not loading on port 3000

  • Cause: wrong port mapping or non-host networking
  • Fix: ensure -p 3000:3000 OR --network host + open browser on correct host IP (make sure using our default template for Flowise and the Secure URL provided in the Server Dashboard)

2. Cannot reach Ollama / “connection refused”

  • Cause: localhost used inside container
  • Fix: use http://host.docker.internal:11434 if Flowise runs in Docker bridge mode

3. RAG not returning documents

  • Cause: missing embeddings or wrong vector DB collection name
  • Fix: verify embeddings output + correct collection name inside Qdrant/Chroma

4. Flow errors: “No LLM model found”

  • Cause: wrong model string
  • Fix: model name must exactly match name in Ollama/OpenAI/etc.

5. Cannot save or load flows

  • Cause: missing persistent volume
  • Fix: mount host path → /root/.flowise

6. GPU model not used

  • Cause: Flowise itself doesn’t run models
  • Fix: GPU must be supported by backend engine (Ollama/comfyui/OpenAI local wrapper)

7. Agent loop / hallucination

  • Cause: high temperature or agent misconfiguration
  • Fix: temperature 0.2–0.4 + enable guardrails + limit tools

8. Wrong output format from tools

  • Cause: tool node output type mismatch
  • Fix: insert “text parser” or “JSON parser” nodes between steps

Calling the ComfyUI API from Flowise

Flowise can call ComfyUI’s API. Install ComfyUI with our tempalte: ComfyUI Template for GPU Server Blibs easily and securely. No native Flowise node yet, but integrate via the HTTP Request Node.

Example minimal setup:

ComfyUI API endpoint

Code 
POST http://<host>:8188/prompt

Flowise HTTP Node fields:

  • Method: POST
  • URL: http://host.docker.internal:8188/prompt
  • Headers: Content-Type: application/json
  • Body (example):
json 
{
  "prompt": {
    "1": {
      "inputs": {
        "seed": 1,
        "steps": 20,
        "cfg": 8,
        "sampler_name": "euler",
        "scheduler": "normal",
        "positive": "a robot with wings",
        "negative": ""
      },
      "class_type": "KSampler"
    }
  }
}

Output handling in Flowise

  • add JSON Parse node
  • feed response into chatbot or UI

Pro tips

Use trigger nodes

Schedule image generations or LLM workflows on intervals/events.

Store generated images

Save ComfyUI output URLs into Qdrant / PostgreSQL for retrieval inside flows.

Async mode

ComfyUI queues jobs → Flowise needs polling logic via loop + HTTP Request node.