Flowise
Building an AI agent or RAG pipeline by dragging nodes onto a canvas instead of writing the glue code.
n8n connects services to each other on a canvas. Flowise does the same thing one layer up: the nodes are LLMs, prompts, retrievers, vector stores, tools, and memory, and what you build is an agent or a RAG pipeline rather than an integration.
It’s the answer to “I want a chatbot over our documentation, but not to write LangChain from scratch.”
Start it
npm install -g flowise
npx flowise start
Then http://localhost:3000. There’s a docker compose up -d path in the repository’s
docker/ folder if you’d rather it live in a container.
What you actually build
| Flow | The nodes in it |
|---|---|
| Docs chatbot | Document loader → text splitter → embeddings → vector store → retrieval QA chain |
| Support agent | Chat model + tools (search, HTTP request, calculator) + conversation memory |
| Data extraction | File input → LLM with a structured-output parser → webhook |
| Multi-step agent | Several agents in sequence, each with its own tools and system prompt |
Every finished flow gets an API endpoint and an embeddable chat widget, which is the part that makes it more than a toy: the canvas is the authoring surface, but the result is something your app can call.
Where it helps and where it doesn’t
It helps when you’re exploring — swapping a vector store, trying a different chunk size, seeing which prompt actually works. Seeing the pipeline as a diagram makes the “where did this answer come from” question answerable, and non-programmers on the team can follow it.
It gets in the way once the thing is real. A canvas diffs badly in code review, version control is exports rather than pull requests, and the moment you need behaviour a node doesn’t offer, you’re writing custom code anyway — at which point the same pipeline in fifty lines of Python is easier to test and deploy. Prototype here; port what survives.
Two practicalities
- Licence is mostly Apache 2.0, with a commercial carve-out. The
packages/server/src/enterprisedirectory and some individually marked files are under a commercial licence. Self-hosting for your own work is fine; read it before you build a product on top. - Keys live in the flow. A Flowise instance holds credentials for every model and service its nodes touch. Don’t leave it on an open port, and give it its own keys with their own limits.
Next
For a narrower automation that publishes on a schedule → Ddalkkak Threads Factory
Last updated 09 Sep 2026, 00:00 UTC.