Flowise — Quick reference

Build AI agents visually — drag & connect "nodes" on a canvas, no code · open-source (Apache 2.0) or Flowise Cloud · strong for RAG over your own PDFs

At a glance

You build an agent by dragging boxes called nodes onto a canvas and connecting them — one node for a language model, one for your documents, one for a tool — so you literally see how information flows. Build and test in the browser, then publish as an embeddable widget or an API. The builder is free; you bring (and pay for) your own LLM API key. Start at flowiseai.com → New Chatflow.

Install & run

Cloud

No install

Open flowiseai.com and start a free Flowise Cloud account — no credit card to begin. Build and test entirely in the browser.

Self-host · npm

Node.js

Install globally, then start the server.

npm install -g flowise
npx flowise start

Open http://localhost:3000

Self-host · Docker

Container

docker run -d --name flowise \
  -p 3000:3000 flowise

Or, from the repo's docker folder: docker compose up -d → http://localhost:3000

Self-hosting the open-source version is free and can run fully air-gapped for sensitive data. It does mean installing/running software (Docker or Node.js).

The three builder types

Type	What it is — when to use
Assistant	The simplest builder — a saved, named assistant with a chosen model and instructions.
Chatflow	One agent with tool calling + RAG together. The everyday choice: gives a single assistant documents and tools and lets it decide when to use each.
Agentflow	Multi-agent orchestration — each node is a step or an agent and the connections set the order. Use when one agent isn't enough.

Core building blocks — the node types

Node

Chat Model

The agent's "brain" (e.g. an OpenAI chat node). Paste in your own model API key. 100+ LLMs supported.

Node

Document Loader

e.g. a PDF File loader — upload a paper or lab notes to ground answers in your own documents.

Node

Vector Store

Indexes your document so the model can retrieve relevant passages before answering — this is RAG.

Node

Tool

Lets the agent take an action (e.g. web search). 100+ integrations across LLMs, embeddings, and vector databases.

Node · Agentflow

Start

The entry point of an Agentflow run — work flows out from here to the agents you connect.

Node · Agentflow

Agents & steps

Specialised nodes — e.g. Search Agent, Get Request, Summarize — wired in sequence so you see the path the task takes.

Build your first flow — a RAG PDF Q&A bot

Step	Do this
`1`	Create a New Chatflow — you get a blank canvas.
`2`	Drag in a PDF File loader and upload one paper or set of lab notes.
`3`	Add a vector store node and connect it (this is the RAG index).
`4`	Connect an OpenAI chat node and paste your own model API key.
`5`	Open the chat panel and ask:

Summarize the main method in this document.

You'll get an answer drawn from passages in your uploaded PDF — a document-grounded bot, not a generic chatbot.

Connect a model, tools & a human

Model

Bring your own key

Add a chat-model node and paste your provider key. The builder is free; model usage is billed by your provider, not Flowise.

Tools

Give it an action

In a Chatflow, connect a tool node to the agent. It decides when to call the tool vs. answer from your documents.

Human-in-the-loop

Pause for sign-off

Add a review step before a consequential action. The run pauses with Proceed / Reject buttons + a feedback box on the last message.

Observe

Traces & evals

Open the execution trace to see each step's input, duration & token count. Use Evaluations & Metrics (on every plan, incl. Free) to score runs.

Deploy / export / API

A finished flow ships as an embeddable chat widget (paste a snippet into a site) or is called via API, CLI, and TS/Python SDKs. The free Cloud plan even allows custom embedded-chatbot branding.

Call a flow over REST — POST /api/v1/prediction/{chatflowId}:

curl -X POST "http://localhost:3000/api/v1/prediction/your-chatflow-id" \
  -H "Content-Type: application/json" \
  -d '{
    "question": "What is artificial intelligence?",
    "streaming": false
  }'

Python SDK

# install
pip install flowise

TypeScript / JS SDK

// install
npm install flowise-sdk

Run it on Flowise Cloud (managed) or self-host the open-source version — including fully air-gapped.

Flowise Cloud pricing

Plan	What you get
Free · $0/mo	2 flows & assistants · 100 predictions/mo · 5 MB storage · Evaluations & Metrics included
Starter · $35/mo	10,000 predictions/mo
Pro · $65/mo	50,000 predictions/mo
Self-host	Open-source (Apache 2.0) — free

On every plan you separately pay your own LLM (and embedding) API usage. Re-check the live pricing page before relying on a number.

Tips

Pick the right builder: a Chatflow (one agent + tools + RAG) handles most tasks — only reach for an Agentflow when one agent genuinely isn't enough. Every agent step in an Agentflow is its own model call, so a multi-agent run spends more predictions and tokens — plan the flow before you run it repeatedly, and read the per-step token counts in the trace to find slow or expensive steps. RAG is Flowise's sweet spot: drop in a PDF + a vector store and your own documents become an answerable knowledge base. Add human-in-the-loop before anything irreversible (sending an email, finalising an answer) so a person signs off first.