Dify vs n8n vs Flowise vs Langflow: Licences, Self-Hosting and Agents (2026)

Dify vs n8n comes down to the job. n8n is workflow automation with AI nodes, under its own Sustainable Use License; Dify is an LLM app builder for chatbots, RAG and agents, under a modified Apache 2.0 licence. Langflow is the MIT-licensed builder for LangChain-style component graphs, and Flowise reached end of life in August 2026, so it is no longer a safe base for new projects.

All four tools put a drag-and-drop canvas in front of language models, so their demos look alike. The differences that last are the licence, who owns the project, what the free self-hosted edition leaves out and whether anyone still maintains it. Each project’s LICENSE file, documentation, release history and pricing page were checked on 14 October 2026.

If you are still weighing automation against agents in general, start with our explainer on agent platforms versus workflow tools. For code-first libraries rather than visual builders, see our guide to open-source agent frameworks.

Which tool fits which job

Licences, ownership and self-hosting

Dify. Commercial use is allowed, including as a backend service or an internal application platform. A commercial licence is needed in two cases: running a multi-tenant environment from the source code without written authorisation, where one tenant equals one workspace, and removing or changing the logo or copyright information in the console or applications while using Dify’s frontend. Contributors accept that LangGenius may make the licence stricter or looser and may use contributed code commercially. Dify’s pricing page describes the free Community edition as a single-workspace edition for open-source enthusiasts, individual developers and non-commercial projects, a narrower framing than the licence text, so raise it with Dify if you plan production use. Kubernetes deployment for high availability, SSO and multi-workspace management belong to Dify Enterprise.

n8n. The Sustainable Use License lets you use or modify n8n only for your own internal business purposes or for non-commercial or personal use, and lets you distribute it only free of charge for non-commercial purposes. Source files with .ee in their name or folder fall under the n8n Enterprise License, which requires a licence for production use but not for development and testing. Without a licence key n8n runs as the free Community edition; registering an email address adds folders, debug-in-editor and custom execution data. n8n’s documentation says npm installation is deprecated from n8n 3.0, so plan on Docker.

Flowise. Workday announced on 14 August 2025 that it had acquired Flowise. On 29 July 2026 the Flowise team announced that it would wind down operations: active development and pull-request reviews stopped that day, the GitHub repository was archived in August, and official presence on Discord and GitHub ended at end of life on 31 August 2026. The team gave the reason as developers moving to coding agents, with rigid low-code workflows hitting limits as tasks grow complex. The Apache 2.0 code stays available to fork. Single sign-on and role-based access sit in the separately licensed enterprise directory, whose production use requires a commercial subscription.

Langflow. IBM announced on 25 February 2025 that it would buy DataStax, which it described as the creator of Langflow, and said it would keep supporting the Langflow community. IBM’s 2025 annual report lists DataStax among the deals it closed that year. Langflow ships as a Python package, a Docker image and Langflow Desktop for Windows and macOS. Read its security page before sharing one instance between teams: the editor runs arbitrary Python with access to the host process, filesystem and network, Langflow does not isolate users within one process, and its user access controls are “designed for usability, not security enforcement”. Authentication and authorisation have to be enforced outside the container, and tenants isolated at the process, disk, network and database level.

Dify vs n8n: LLM app builder or workflow automation

The two products start from different units of work.

In n8n the unit is a workflow: a trigger such as a webhook, schedule or app event runs a chain of nodes that move data between systems. AI arrives as nodes in that chain. The AI Agent node reasons and calls tools, vector store nodes insert and retrieve documents, the MCP Client Tool node lets an agent use external MCP servers, and the MCP Server Trigger exposes n8n workflows to MCP clients over SSE or streamable HTTP, though not stdio. Cloud plans are priced by workflow executions, and every plan includes unlimited users.

In Dify the unit is an AI application: a chat app, an agent or a workflow built around a model, published as a web app, an API, an embed or an MCP server. Knowledge bases with configurable chunking and retrieval come built in, and knowledge pipelines handle custom ingestion. Dify can also use external MCP servers as tools, over HTTP transport only, with OAuth handled for you. Hosted plans are priced per workspace and include message credits for hosted models.

Choose n8n when the language model is one step in a business process that touches many systems. Choose Dify when the AI application is the product people use. The two can also coexist, with n8n handling the business steps around an assistant built in Dify. The licences differ too: n8n’s free edition is limited to internal or non-commercial use, while Dify’s allows commercial use but restricts multi-tenant hosting and branding changes.

Flowise vs Langflow after Flowise’s end of life

Until mid-2026 these two were the obvious pair: visual canvases for chaining model calls, retrieval and tools. Flowise offered three builders, Assistant, Chatflow and Agentflow, on a node library organised around LangChain components plus LlamaIndex, with MCP client and server nodes. That comparison is now mostly about migration.

If you run Flowise today, you have three options:

1. Pin and freeze. Keep the last release, 3.1.4, running inside your network. The Apache 2.0 licence allows it, but every future security fix is yours to write.
2. Fork. The Flowise team itself encourages internal or community-led forks. That works for teams with JavaScript skills and a reason to keep the Flowise data model.
3. Migrate. Langflow is the closest in shape: a visual editor, agents, flows served as APIs, and MCP support in both directions. It is written in Python rather than Node.js, and we found no documented converter for Flowise flows, so treat the move as a rebuild.

Langflow is also actively maintained, with a patch release every week through September and October 2026. Before you move several teams onto one instance, plan the identity and isolation layer described in the security notes above.

Langflow vs n8n and Dify vs Langflow

Langflow vs n8n. They sit at different layers. Langflow builds the AI logic itself: components you can open and edit in Python, wired into flows that run behind the Langflow API or as MCP tools. n8n automates the business process around such logic, with schedules, triggers and integrations across many SaaS applications. Langflow is MIT-licensed; n8n’s free edition is limited to internal use. For a team of developers building retrieval or agent services, Langflow is the natural canvas. For operations teams wiring an AI step into an approval flow or a CRM update, n8n is.

Dify vs Langflow. Both are LLM application builders, aimed at different users. Dify packages each application with a ready-made web interface, a knowledge-base UI and four workspace roles: owner, admin, editor and a normal role that can only use published apps. Langflow gives developers source-level control of every component, MIT licensing and an MCP server for every project, and expects you to bring your own authentication. If non-technical staff will build and run assistants, Dify’s packaging helps; if engineers will build services that other systems call, Langflow’s openness helps.

Agents, RAG, MCP and team features

Identity is where the free editions differ most.

Two details deserve attention in a security review. Dify’s published MCP server URL contains its own credentials, so the documentation tells you to treat it like an API key. Langflow seeds viewer, developer and admin roles, but in the open-source build those roles restrict nothing until an enforcement plugin is registered.

Cloud plans and published prices

The units are hard to compare directly. A busy workflow can spend n8n executions quickly, while a Dify deployment with many departments multiplies workspaces. n8n also offers qualifying companies with fewer than 20 employees a start-up plan at half the Business price. For a deeper look at what Dify costs once it is self-hosted, see our Dify cost breakdown.

Where LangChain, LangGraph, CrewAI, Langfuse and Coze fit

Searches pair these four builders with names that belong to other layers:

• LangChain and LangGraph are MIT-licensed code libraries, not visual builders. n8n ships its AI nodes in a package named after LangChain, but a LangGraph agent lives in your own codebase and runtime.
• CrewAI is an MIT-licensed Python framework for orchestrating role-playing, autonomous agents in code.
• LangSmith and Langfuse trace and evaluate model applications rather than build them. LangSmith is LangChain’s commercial platform, free for one seat, $39 per seat a month on Plus, with self-hosting reserved for Enterprise. Langfuse is MIT-licensed outside its enterprise directories. Langflow’s documentation lists both as observability integrations.
• Coze Studio is the open-sourced core of the Coze agent development platform, under Apache 2.0 with a Go backend. It covers much the same ground as Dify: prompts, retrieval, plugins and workflows.

For the open-source versus commercial trade-off across this whole category, see our comparison of open-source and commercial agent platforms.

How VDF AI fits

All four builders leave the governance layer to whoever runs them: identity, permissions per resource, audit evidence and model policy. That is fine for a team’s own projects and becomes the main cost once several departments and regulated data are involved.

VDF AI is the governed on-premises alternative for that stage. It provides governed agents and multi-step workflows, private RAG over company data and audit logs, with role-based access control on every plan, native Microsoft Entra ID single sign-on on-premises, and other identity providers through an SSO-aware reverse proxy. Model choice runs through VDF AI Router, which can register an Ollama endpoint or a custom on-premises deployment next to the cloud models you approve; in air-gap mode, every request stays on the local ones. On-premises, VDF AI is licensed in seat bundles and model tokens are not metered; see current pricing.

Our head-to-head pages compare VDF AI with n8n and Flowise, and the self-hosted alternative pages cover replacing n8n for agent workflows and moving off Flowise.