Flowise vs n8n: LLM chains or process automation

If you want to build something with a model and no code, two tools tend to show up: Flowise and n8n. They look alike, with the same blocks and arrows on a canvas, but they solve different problems. The "flowise vs n8n" question is best settled by what you are running: a dialogue with a user or a background process. Below we show how to self-host each, how to connect a model through an OpenAI-compatible API and roughly what it costs.

Flowise vs n8n: what differs

Flowise's documentation calls it an open-source platform for building AI agents and LLM workflows. It has three visual builders: Assistant (the easiest way to get an assistant), Chatflow (chatbots and single agents, chains with retrieval, reranking and similar techniques) and Agentflow (multi-agent systems and complex orchestration). Flowise nodes mirror LangChain and LlamaIndex components: chat models, vector stores, memory, agents. A finished flow is embedded on a site as a widget or called through an API.

n8n describes itself as a fair-code licensed workflow automation tool with AI features added. Its strength is the trigger: a schedule, a webhook, an event in a third-party service. The model is one step in it, with AI Agent and Basic LLM Chain nodes available.

A rough rule: if a flow has someone to talk to, look at Flowise. If a flow has a schedule or an event, look at n8n.

Self-hosting

Both can be used in the cloud, but self-hosting is the question people ask most, so here is the minimum.

Flowise needs Node.js (the project README asks for version 20.0.0 or newer):

npm install -g flowise
npx flowise start

The interface opens at http://localhost:3000. With Docker: the project repository has a docker folder, where you copy .env.example to .env and run docker compose up -d.

n8n is best installed through Docker Compose, the method recommended in their documentation. A quick check with a single container:

docker volume create n8n_data
docker run -it --rm --name n8n -p 5678:5678 \
  -e GENERIC_TIMEZONE="Europe/Moscow" -e TZ="Europe/Moscow" \
  -v n8n_data:/home/node/.n8n n8nio/n8n

The interface is at http://localhost:5678. Mind your data: the n8n_data volume holds your workflows and credentials. The n8n documentation warns that self-hosting assumes server experience and that mistakes can lead to data loss and security problems. The same holds for Flowise: if you expose it to the internet, put authentication in front and keep it updated.

Connecting a model

You need a fetch_... key from the dashboard↗ and the address https://api.zukko.pro/v1. The request format is in the chat/completions reference↗.

Flowise

1. On the canvas, take the ChatOpenAI node from the Chat Models group.
2. Under Connect Credential, create credentials and paste fetch_....
3. Open Additional Parameters and set Base Path to https://api.zukko.pro/v1. The Flowise documentation describes this approach for all OpenAI-compatible services.
4. In Model Name, type an ID such as gpt-6.1-sol.

For OpenAI-compatible services Flowise also has a separate ChatLocalAI node, which also takes a Base Path. If Flowise runs in Docker and a local model runs on the host, host.docker.internal can replace localhost; on Linux the documentation suggests the Docker gateway address. With our gateway this is not needed, since it is reachable from the internet. More in Flowise and Zukko↗.

n8n

Create OpenAI credentials with the fetch_... key. If your version has a Base URL field in them, set https://api.zukko.pro/v1: the OpenAI Chat Model node then goes through the gateway. If there is no such field or the model list does not load, use the HTTP Request node: POST to https://api.zukko.pro/v1/chat/completions, header Authorization: Bearer fetch_..., with model and messages in the body. For OpenRouter users, moving means changing the address, the key and model names; the steps are in Moving from OpenRouter↗, and a general n8n example is in n8n: automation with Claude and GPT↗.

Two examples

A FAQ widget (Flowise). A Chatflow: documents with questions and answers, retrieval over them, a chat model and conversation memory. The result is embedded on a site. The load is unpredictable: visitors arrive, not a schedule. Dialogue history and spend limits matter here.

A nightly news digest (n8n). A schedule trigger, an RSS request, a loop over entries, the model summarizes each, the result goes to Telegram. The load is predictable: the same number of calls every night.

What it costs

Prices as of October 18, 2026, per 1M tokens.

Widget: 3,000 replies a month, each with 3,000 input tokens (history and retrieved fragments) and 300 output tokens, model claude-sonnet-5-5 (input 0.4, output 2). Per reply: 3,000 × 0.4 + 300 × 2 = 1,200 + 600 = 1,800 micro-dollars. Per month: 1,800 × 3,000 = 5,400,000 micro-dollars, which is $5.40. This is a no-cache estimate in the OpenAI format.

Digest: 200 entries a day, 1,500 input tokens and 150 output tokens each, so 300,000 input and 30,000 output tokens a day.

• glm-5.3-flash (0.04 / 0.04): 0.3 × 0.04 + 0.03 × 0.04 = 0.012 + 0.0012 = $0.0132 a day, $0.396 over 30 days.
• gpt-6.1-sol (0.27 / 1.32): 0.3 × 0.27 + 0.03 × 1.32 = 0.081 + 0.0396 = $0.1206 a day, $3.618 over 30 days.

Whether a cheap model is enough for a short summary is shown by a test on your material. Model list and prices: GET https://api.zukko.pro/v1/models. For an automated process, give the key a daily dollar limit: if the flow starts looping, it hits the limit and gets a 429 instead of draining the balance. Top up through SBP (the payment page adds an 8.5% fee on top), cryptocurrency or MasterCard.

What to choose

• You need a site chat or an internal assistant with memory and a document base: Flowise.
• You need to connect services and fire on an event or a schedule: n8n.
• You need both: install both. Flowise exposes a finished flow through an API and n8n calls it with an HTTP Request node.
• You do not want to administer anything: use the cloud versions and check the project sites for terms.

FAQ

Are Flowise and n8n free to use?

Both can be self-hosted, but you pay for the server and for model tokens. Check the project sites for license and cloud plan terms; we do not quote them.

Which model do agents need?

Agents need tool calling. Test it on your model with one scenario, then choose by price and context. claude-sonnet-5-5 or gpt-6.1-sol is a fair start.

Why does Flowise complain about the model or key?

Most often the Base Path is wrong: it needs https://api.zukko.pro/v1 with the suffix. If you get 401, the key has a stray space or is not a fetch_... key.

Where do I see spend?

In the "Usage" section of the dashboard and through the balance and usage history API; links are in the documentation.