Short answer. For a new AI workflow, choose n8n. It is actively maintained, prices by execution rather than step, and connects to the business systems around the model. Flowise is no longer a safe new dependency: the company announced its wind-down on 29 July, 2026, archived the repository on 10 August 2026 and ended official support on 31 August 2026. The code is Apache 2.0 and can be forked, but you would own maintenance.
Side by side
| Dimension | Flowise | n8n |
|---|---|---|
| Category | Visual LLM app builder | Workflow automation platform |
| Status | Winding down; repository archived | Active development |
| Core object | Assistant, chatflow, agentflow | Workflow, node, trigger, credential |
| RAG | Built-in knowledge bases and retrievers | Assembled from AI and vector store nodes |
| Integrations | 100+ models, tools and vector stores | Hundreds of business-system connectors |
| Self-host | Apache 2.0; Docker | Fair-code Community Edition; Docker and queue mode |
| Licence | Apache 2.0 | Sustainable Use License for the Community Edition |
| Cloud entry | $35/month Starter (legacy page) | 20 euros/month Starter, annual billing |
| Best for | Existing deployments and forks | New automation and AI agent workflows |
Sources: Flowise sunset notice and docs, n8n pricing page, checked 20 November 2026. Flowise's cloud pricing page may remain online after the wind-down; availability is not guaranteed.
What changed in 2026
Flowise's own notice is unusually direct. The team said developers were moving to coding agents for complex work and that rigid low-code workflows hit a ceiling as tasks grew. Feature development stopped at the announcement, pull requests were closed, the GitHub repository moved to public archive, and npm and Docker images were marked deprecated.
That does not make existing Flowise deployments disappear. The code still runs, and an Apache 2.0 fork is legally possible. It does change the risk calculation. An archived project receives no security patches, no dependency updates and no compatibility fixes when a model provider changes an API. If the workflow touches customer data, that is an operational liability, not just an inconvenience.
n8n is on the other side of that curve. It has a commercial vendor, a community edition, a cloud service and an enterprise tier. Its AI features are younger than Flowise's LLM-specific builder, but the platform is maintained.
Visual builder vs automation
Flowise organised the canvas around the LLM application: a chatflow starts with a chat trigger, retrieves from a knowledge base, calls a model and returns an answer. Agentflow extended that to multi-agent orchestration. The mental model is "build the AI feature", and the output is an API or embedded chat widget.
n8n organises the canvas around a business process: a webhook, schedule or app event starts the workflow, data is transformed, decisions branch, and side effects happen in a CRM, ticketing system or database. The AI Agent node is one step in that process. The mental model is "automate the work around the model", which is why n8n has deeper connectors and retry/error handling.
A chatbot over documents fits Flowise's shape; a ticket-triage process with approvals fits n8n's.
LLM nodes and RAG
Flowise packaged RAG: document loaders, text splitters, embeddings, vector store connectors, retrievers and citations were nodes you could wire together, and the knowledge base feature handled ingestion for simpler apps. It also shipped evaluations, tracing and human-in-the-loop review.
n8n provides the same ingredients as separate nodes: document loaders, text splitters, embedding models, vector store nodes and an AI Agent with tools. You choose the chunking strategy, the retrieval parameters and the evaluation method. That is more work, but it is also easier to inspect and version as code.
For either tool, the model itself can be any OpenAI-compatible endpoint. Point the credential at https://api.plugsky.com/v1 and use a model ID such as plugsky-chat or plugsky-embed. The workflow engine does not need to know which vendor serves the tokens.
Self-hosting and limits
Flowise self-hosts with Docker and stores data in Postgres plus a vector store and object storage. It scales horizontally with a message queue and workers. Because the project is archived, you should pin versions, mirror images and review the dependency tree before exposing it to the internet.
n8n self-hosts with Docker or Docker Compose; queue mode adds Redis and workers for concurrency. Cloud plans cap executions rather than steps: Starter includes 5 concurrent executions and 2,500 workflow executions per month; Pro raises concurrency and history; Business (self-hosted) adds SSO, environments, Git control and queue mode. Execution time limits also differ by plan, so a long-running agent loop may need Pro or Business.
Pricing
n8n Cloud's public prices are execution-based. Starter is 20 euros per month billed annually for 2,500 executions; Pro is 50 euros for 10,000; Business is 667 euros for 40,000 and is self-hosted only. All plans include unlimited users and workflows. Extra executions are capped rather than billed without limit. The Community Edition is free to self-host, and enterprise features are licensed.
Flowise's cloud page still listed Starter at $35 per month and Pro at $65 plus $15 per user. Those prices are historical. With the company winding down, budget for migration or for the engineering time to maintain a fork.
Model tokens are separate in both cases. A flat workflow plan does not include inference, so the model bill grows with traffic even when the automation plan does not.
Migrating off Flowise
Inventory every chatflow and agentflow, then classify it: simple RAG, tool-using agent, or multi-step orchestration. Export prompts, knowledge base documents and evaluation sets. Rebuild the simple RAG flows first in n8n; they map cleanly to a trigger, a retriever, a model call and a response.
# Flowise prediction call (legacy shape)
curl -X POST "https://your-flowise-host/api/v1/prediction/FLOW_ID" \
-H "Content-Type: application/json" \
-d '{"question":"What is the refund window?","overrideConfig":{"sessionId":"u-42"}}'
# n8n: expose a webhook, then call it from the same application.
curl -X POST "https://your-n8n-host/webhook/refund-answer" \
-H "Content-Type: application/json" \
-d '{"question":"What is the refund window?","userId":"u-42"}'
Keep the old endpoint alive behind a proxy while you compare answers. Log retrieved chunk IDs, prompt version and model ID on both sides; otherwise a regression is impossible to attribute. Retire Flowise only after the evaluation set passes.
FAQ
Is Flowise shutting down?
Yes. Flowise announced a wind-down on 29 July, 2026, archived its GitHub repository on 10 August 2026 and ended official support on 31 August 2026. The Apache 2.0 code remains available to fork, but there are no new official releases.
Should I choose Flowise or n8n for a new AI workflow?
For a new project, choose n8n. Flowise is no longer actively maintained, so a new deployment carries maintenance and security risk unless you are prepared to own a fork.
Can n8n build a RAG chatbot?
Yes. n8n has AI Agent nodes, vector store nodes, embeddings and document loaders. You assemble chunking, retrieval and evaluation yourself, whereas Flowise packaged those steps into its chatflow and knowledge base features.
How does n8n pricing work?
n8n Cloud prices by workflow executions, not steps or users: Starter is 20 euros per month billed annually for 2,500 executions, Pro is 50 euros for 10,000, and Business is 667 euros for 40,000 on self-hosted deployments. The Community Edition is free to self-host.
Can I keep running Flowise?
You can run the archived code indefinitely, but you own patching, dependency upgrades and security fixes. Before staying, confirm that your fork has an owner, a release process and a plan to replace unsupported dependencies.
Reviewed 20 November 2026. Flowise's status was checked against its sunset notice; n8n prices and limits were checked against the n8n pricing page and n8n docs. Flowise cloud pricing is historical and should not be used for new commitments.