Flowise vs n8n: one is archived, the other still ships

Flowise vs n8n is no longer a feature comparison: FlowiseAI archived the Flowise repository on August 13, 2026 and ended core-team support on August 31, 2026, while n8n shipped a stable release on September 21, 2026. If you are choosing today, n8n is the maintained option. If you already run Flowise, the honest answer is narrower — your existing flows still run, the Apache-2.0 code is still yours, and moving them is a rebuild rather than an import, because neither vendor documents a converter.

One boundary before anything else. The vendor's sunset notice scopes every milestone it lists to a column headed "Impact on OSS Repository". It says nothing about Flowise Cloud, and on September 19, 2026 flowiseai.com still rendered paid plans behind a "We're sunsetting Flowise" banner. This page does not claim the hosted service is dead, and it does not tell you it is safe to buy. The vendor has published nothing either way.

Project status: the difference that decides this

Checked September 19, 2026 against the Flowise repository, the n8n repository, Flowise's LICENSE.md, n8n's LICENSE.md, the npm entry for n8n — re-read September 21, 2026, when it served 2.39.9 as both latest and stable, published that day — and the Workday acquisition release. That release praises the project's open-source foundation and says nothing about the 2026 sunset; the two documents do not connect, so do not read the acquisition as the cause.

Four loose ends worth knowing, because each one will mislead you if you meet it alone. First, the announced archival date and the real one differ: the notice scheduled August 10, 2026, the GitHub banner reads August 13, 2026. Second, the notice said npm packages and Docker images would be marked deprecated at archival — on September 21, 2026 the npm registry still served flowise@3.1.4 as latest with no deprecation marker on any published version. Docker Hub was not checked. Third, docs.flowiseai.com is still live, carries no end-of-life banner, and still opens by calling Flowise an open-source generative AI development platform, so a reader who lands there gets no warning. Fourth, the licence is not MIT, so any roundup describing it that way is wrong; read the LICENSE.md above.

Who should pick which

Asked as n8n vs Flowise — someone already on n8n wondering what Flowise offered — the answer is short: a purpose-built visual builder for LLM chains and agents, which n8n covers with an AI node family instead, and which is now frozen. There is no reason to adopt Flowise from n8n in September 2026. Asked the other way round, it is a real decision with five parts.

Pick n8n if the workflow has to keep changing. New model names, new integrations, new protocol support and security fixes will only arrive on one side of this comparison. n8n is also the broader tool: LLM nodes are one family inside a general automation platform, so the same workflow can carry the API calls, branching, scheduling and error handling around the model call. That breadth is the cost too — if all you want is a retrieval chatbot, you are adopting a much larger product to get it.

Pick n8n if permissions and execution shape matter. Its metering unit is defined and generous in the right direction: one run of a whole workflow is one execution regardless of step count. Its self-hosted Community edition is free with no documented execution or workflow cap, and registering an email adds folders, debug-in-editor and custom execution data at no cost. What it holds back is a specific, published list — projects, SSO, environments, external secrets, Git version control, workflow and credential sharing, log streaming, multi-main mode, custom variables and external binary storage all require a paid plan or licence key, on Cloud or self-hosted alike (n8n docs, checked September 19, 2026). If your team needs SSO or per-project separation, n8n is a paid product, not a free one.

Stay on Flowise, for now, if the flow is finished and self-hosted. An end-of-life date does not reach into a self-hosted install: the Apache-2.0 code keeps running, the pinned container keeps starting, and the sunset notice explicitly leaves the code for teams to keep building on — it encourages forking the repo for internal updates or community-led forks. The honest version of "stay" is that you are now the maintainer. No upstream patch is coming, including security patches, and no fork has been surveyed here as a credible successor — this page names none.

Leave Flowise now if you are on a paid hosted plan or depend on new models. Those two carry live risk rather than theoretical risk. The hosted service's future is unstated by the vendor, and Flowise's model dropdown is loaded from a file in the archived repo, so the list of names it offers is frozen — the workaround exists but it is now permanently a workaround.

Migration cost is the fifth factor, and it is high. No importer for the other product is documented by either vendor: n8n's export-and-import page covers n8n workflow JSON from a URL or a file and names no other platform, and nothing in Flowise's documentation offers an n8n path. With no documented import path, credentials do not travel either — plan on re-entering every API key by hand on the new side. Count nodes, not flows, when you estimate the work.

Cost model and capability, side by side

Published plans, and why you cannot just add them up

Both tables read September 19, 2026, from flowiseai.com and n8n.io/pricing. Three cautions. The currencies genuinely differ as presented: n8n rendered in EUR on every fetch made here, and no USD presentment was confirmed, so do not put 20€ and $35 in the same sum without a rate you have sourced yourself. The units are not convertible either — an n8n execution is one whole workflow run by the vendor's own definition, while a Flowise prediction is undefined on any page found during this check. And n8n's own docs note that the per-plan feature split changes, so confirm in checkout rather than from this table. Third-party hosting resellers publish their own "n8n pricing"; only n8n.io is authoritative.

One managed option deliberately carries no number here. n8n Cloud offers Gateway credits as an alternative to bringing your own model key, but the docs put the rates behind a login, so none is published on this page. What the docs do state is the shape: Starter and Pro only, n8n 2.36.0 or later, not available on Cloud Enterprise or on any self-hosted install, and top-up credits expire twelve months after purchase. That last line matters for this comparison — self-hosting is exactly where most Flowise refugees land, and it is exactly where Gateway credits do not reach. Bringing your own key is the path that reaches both deployments, self-hosted included.

Pointing either one at your own key

This is where the two products stop resembling each other, and the details below were read from source on September 19, 2026 rather than from marketing pages.

In n8n, the override is a credential field, and the editions page does not list it as a paid feature. The OpenAI credential carries Base URL, defaulting to https://api.openai.com/v1, alongside the API key and an optional custom header pair. The Anthropic credential has the same field, defaulting to https://api.anthropic.com, sends the key as x-api-key, and tests with a GET /v1/models carrying anthropic-version: 2023-06-01. Two practical consequences. First, each credential test fetches a model list — {Base URL}/models on the OpenAI credential, {Base URL}/v1/models on the Anthropic one — so an endpoint that does not answer a model list will show a red credential even while chat works. Second, the node-level Base URL option is hidden for node typeVersion 1.1 and above in the OpenAI Chat Model node — if you are following an older tutorial that says to set it on the node, that field will not be there.

Model ids are the easy part on this side. The picker is a resource locator with a "From List" mode that queries your endpoint live and an "ID" mode that takes free text, so a gateway's own model names work without patching anything. n8n also shows an in-product warning you should take literally: when a Base URL override is set, it says that not all models might be chat-compatible or support other features like tool calling or JSON response format. Treat that as the vendor telling you to test your own flow.

One toggle deserves a check rather than an assumption. The same node can send either Chat Completions or the Responses API. n8n's documentation says the node defaults to Chat Completions unless you switch on "Use Responses API", while the master-branch source declares responsesApiEnabled with default: true for the current maximum typeVersion. The two disagree and this page could not resolve it without running n8n, so open the node and look at the toggle before you debug anything else. It is load-bearing: the Responses-only built-in tools — Web Search, File Search, Code Interpreter — run inside the provider's own Responses loop and only in combination with the AI Agent node, and that is the one genuine capability a third-party endpoint can take away from you.

In Flowise, the override is a node parameter and the model list is frozen. Its OpenAI credential holds exactly one field, the API key — there is no Base URL on it. The endpoint lives on the ChatOpenAI node under Additional Parameters, as Base Path plus a Base Options JSON blob for extra headers; Flowise's own ChatOpenAI page names TogetherAI and OpenRouter as the worked examples for exactly that. The node's Model Name, however, is a dropdown rather than free text, and its loader reads MODEL_LIST_CONFIG_JSON with a fallback to a models.json file inside the archived repository. That file still returned HTTP 200 on September 19, 2026, at 120,125 bytes, listing 47 chatOpenAI entries frozen at the code freeze — a list that can never grow again from upstream. Use the separate ChatOpenAI Custom node, whose Model Name is a plain string and which also has Base Path, or set the env var to a file of your own.

And one gap worth planning around: Flowise's ChatAnthropic node has no Base Path or Base URL field at all in 3.1.4 — every input label on it was enumerated to confirm that, and its init passes no endpoint override to the client either. So there is no in-product way to point that node at a third party; whether the underlying SDK can be redirected some other way was not established here. Plan on the OpenAI-compatible path instead.

For Kunavo specifically, that maps to an OpenAI-compatible base URL of https://api.kunavo.com/v1 with the key as a bearer token, and on n8n's Anthropic credential the origin https://api.kunavo.com with no path, since the client appends /v1/messages itself. Kunavo's Messages endpoint accepts the credential as either x-api-key or Authorization: Bearer, which is the header format n8n's Anthropic credential sends. That is a statement about documented endpoints on both sides, not a compatibility result: Kunavo has run no runtime test against n8n or Flowise, and neither product is listed as officially supported here. Keep a working route available while you try it. See the chat endpoint and base URL reference for the exact values.

A worked token estimate

This is illustrative token arithmetic, not a measured workflow cost and not a bill ceiling. Assume a support-answering flow that runs 2,500 times a month — the count n8n Starter includes — sending 6,000 input tokens and receiving 500 output tokens per run, with no caching and no retries. Rates are live Kunavo catalog prices per million tokens.

The useful reading is the spread, not the winner. Under these assumptions the same flow costs $8.50 on Claude Haiku 4.5 and $25.50 on Claude Sonnet 4.6 — three times the token line for identical workflow logic. Note what that does not show: at this volume the token line and a platform subscription are the same order of magnitude, so neither dominates and the platform decision cannot be made on token price alone. Scale the run count up and the ratio flips, which is the real reason to keep the model choice in your own hands rather than inside a bundled credit balance. Cheapest listed rate and cheapest way to finish the job also remain different questions, and a model that needs a retry can cost more than a pricier one that does not; AI cost optimization covers how to measure that on your own flow.

Two exclusions to keep the arithmetic honest. If your flow does retrieval, the embedding step is not served by Kunavo — Kunavo's catalog has no embedding model, so that part runs locally or against an outside provider and is billed there; RAG implementation covers that split. And Kunavo's catalog amount is a billing floor rather than a cap: when the upstream reports its charge, the bill is the greater of catalog cost and upstream cost times the applicable markup. The minimum top-up is $10 in prepaid credit — a funding minimum, not a task fee or a subscription. See billing details.

If you are moving, move the keys once

The rebuild is the expensive half; the credential layer does not have to be. Both products take one OpenAI-compatible endpoint and one key, so a gateway lets you stand the new n8n flow up beside the old Flowise one on the same balance, compare outputs, and cut over without opening two vendor accounts. On n8n you get the Anthropic path as well, which Flowise's ChatAnthropic node does not expose. The OpenAI-compatible API guide explains what that contract actually covers, and the AI agent API directory lists the other clients that accept the same two fields.

If you are still comparing providers rather than platforms, OpenRouter alternatives and best LLM gateway cover that choice. When you are ready to fund a key, create a Kunavo account, then run one bounded workflow and read the charge your account actually recorded for it.

FAQ

Checked September 19, 2026: both repositories and their LICENSE.md files, the Flowise sunset notice and homepage pricing, the npm registry entries for flowise and n8n, n8n's pricing page, editions doc and Gateway credits doc, and the node and credential source for n8n's OpenAI and Anthropic paths and Flowise's ChatOpenAI, ChatOpenAI Custom, ChatAnthropic and model loader. Not checked: Flowise Cloud's operating status, its refund or data-export policy, n8n Gateway credit rates, any community fork, whether the Flowise Anthropic client can be redirected by any means outside the node UI, Docker Hub deprecation markers, and whether either product works against Kunavo at runtime — no compatibility test was run. Release and npm facts re-read September 21, 2026. Kunavo token rates come from the live catalog and every dollar figure here is illustrative token arithmetic.