n8n vs Langflow: Which Should You Use for AI Workflows?

The n8n vs Langflow choice depends on the main job. n8n is a workflow automation tool that connects business apps and adds AI steps to a process. Langflow is a visual builder for AI apps, such as an agent or a chatbot that answers from your own documents. Use n8n if the work is a business process with triggers, app integrations, and approvals. Use Langflow if the main work is the AI logic, and a Python developer will change it.

You can also use both tools. n8n runs the business process, and it sends one HTTP request to a Langflow flow for the AI step.

This comparison uses the official documentation, GitHub data, price pages, and security records of each tool. I checked all data on 11 November 2026. I did not run a speed or quality test of either tool.

What is the difference between n8n and Langflow?#

The main difference between n8n and Langflow is the start point. n8n starts from a business process. Langflow starts from an AI application.

n8n is a workflow automation platform that runs a series of steps when an event occurs. n8n calls each step a node. A trigger node starts each workflow, for example on a form submission, a schedule, a new email, or a webhook call. A webhook is a URL that receives data from another system. The other nodes call apps such as Slack and Gmail, or call an AI model.

Langflow is an open-source tool in Python that you use to build AI applications in a visual editor. You connect components, such as a model, a prompt, a vector store, and an agent, into a flow. You test the flow in the Playground, which is a chat window in the editor. Other software then calls the flow through the Langflow API or as an MCP tool.

These technical names occur in this guide:

• RAG (retrieval-augmented generation) is a method that finds relevant text in your documents. It gives that text to an AI model before the model answers.
• An embedding is a list of numbers that represents the meaning of a text.
• A vector store is a database that stores embeddings and finds similar items.
• MCP (Model Context Protocol) is a standard that lets an AI app find and call tools on other systems.
• A CRM (customer relationship management system) is an app that keeps the records of your customers and sales.

n8n gives the same summary on its comparison page about Langflow. The page says: "Langflow starts from the AI application, n8n starts from the end-to-end business workflow." n8n wrote that page about a competitor, so check its other claims against the Langflow documentation.

Compare n8n and Langflow side by side#

This table compares n8n and Langflow on the data that I checked on 11 November 2026. Versions and star counts change often.

GitHub stars show interest, not quality. The n8n integration count also changes between n8n pages. The GitHub description says "400+", and the integrations page lists more than 2,000 entries. Compare the integrations that you need, not the totals.

Langflow is part of DataStax. In February 2025, IBM announced a plan to buy DataStax. IBM publishes the Langflow security advisories, such as the advisory for CVE-2026-9198. The US National Vulnerability Database (NVD) lists the product as "IBM Langflow OSS".

How does each tool build an AI agent?#

Both n8n and Langflow give you a visual AI agent that can call tools. An AI agent is an AI model that selects tools, such as a search or an API call, to complete a task. The main difference is what the agent can connect to.

The AI Agent node in n8n#

n8n uses cluster nodes for AI. A cluster node is a root node that has sub-nodes. The AI Agent root node takes sub-nodes for the chat model, the memory, and the tools. Other root nodes are chains and vector stores. n8n builds these nodes on LangChain, which is a library for apps that use large language models (LLMs).

The AI Agent node has three features that matter for business work:

• MCP tools: The MCP Client Tool sub-node lets the agent call the tools on an MCP server.
• Human approval: n8n can ask a person to approve a tool call before the agent runs it. The request can go to Slack, Microsoft Teams, Gmail, Telegram, and other channels.
• Evaluations: n8n can run a workflow against a set of test inputs and give the results a score. The evaluations overview describes the two types.

An agent can also be a step in a larger workflow. The workflow can read a CRM record before the agent runs. After the agent runs, the workflow can write the result to other apps.

The Agent component in Langflow#

Langflow builds an agent from components on a canvas. You connect an Agent component to a model component and to tool components. The MCP Tools component connects the agent to the tools on a local or remote MCP server.

A Python developer can add a custom component. A custom component is a Python class that inherits from Component and declares its inputs and outputs. A team can thus add its own retrieval rules or API calls to the canvas as a component.

Each Langflow project also has its own MCP server. Each flow in the project becomes a tool on that server, if the flow has a Chat Output component. Other MCP clients can then call your flows as tools.

How do triggers and app integrations differ?#

n8n has many more app integrations than Langflow, and it has a documented scheduler. For business automation, this is the largest difference between the two tools.

n8n can start a workflow from a schedule, a webhook, a form, or a chat message. It can also start from an app event, such as a new email in Gmail. After the trigger, the workflow can call the apps of the business process. Examples are a CRM, a help desk, a spreadsheet, and a chat tool.

Langflow 1.13 can start a flow in these ways:

1. An API call to the run endpoint of the flow.
2. The Webhook component. It needs an API key by default, and it returns "Task started in the background", not the output of the flow.
3. An Inbound Webhook trigger, for a system other than Slack, Microsoft, or Google.
4. An app trigger for Slack, Google Workspace, or Microsoft 365. Examples are a new Gmail message, a new Outlook message, and a new file in Google Drive.
5. An MCP client that calls the flow as a tool.

The app triggers need more setup than the n8n triggers. You enable each trigger through the Langflow Triggers API. The Langflow Connection for the app account must belong to the owner of the flow and must allow background runs. Some triggers also need a public HTTPS URL or a separate listener process.

The Langflow 1.13.2 code contains a Schedule trigger component, but the user documentation does not describe it. An open pull request for version 1.14 adds a schedule picker to that trigger. Until Langflow adds the trigger to its documentation, call the flow from cron or from n8n on a schedule. The cron expression generator helps you write the schedule.

Langflow lists 80 component bundles in its documentation. Most bundles connect to model providers, such as OpenAI, Anthropic, and Ollama, or to vector stores, such as Pinecone, Qdrant, and pgvector. Version 1.13 added Connections, which keep accounts for Google Workspace, Microsoft 365, and Slack. I found no bundle for a CRM such as HubSpot or Salesforce.

How does RAG work in n8n and Langflow?#

Both n8n and Langflow can build a basic RAG pipeline on a canvas. A RAG pipeline has two parts:

1. Ingestion loads the documents and cuts them into chunks. A chunk is a short part of a document. Ingestion then makes an embedding for each chunk and stores it in a vector store.
2. Query finds the best chunks for a question and sends them to the model with the question.

In n8n, vector store root nodes, such as Pinecone, Qdrant, and Simple Vector Store, do the storage. Sub-nodes add the embedding model, the document loader, and the text splitter. This setup is good for a RAG step inside a larger workflow.

In Langflow, vector store components, such as Pinecone, Qdrant, Milvus, Weaviate, and pgvector, do the storage. A developer can change each step in Python, and the team can test answers in the Playground. This setup is good when the RAG app itself is the product.

For both tools, a canvas works well for a simple RAG step. If your documents are messy PDFs and tables, or if users must see only some documents, the retrieval work needs code and tests. The guide to LangChain vs LlamaIndex for RAG pipelines compares two code frameworks for that work. The guide to RAG evaluation for accuracy and latency shows how to measure the answers. To select a vector store, read what a vector database is and how to choose one.

How do you call a Langflow flow from n8n?#

To call a Langflow flow from n8n, use an n8n HTTP Request node. The node sends a POST request to the run endpoint of the flow. With this setup, n8n runs the trigger and the apps, and Langflow runs the AI step.

Here is an example of a workflow for support email:

1. A Gmail Trigger node in n8n receives a new support email.
2. An HTTP Request node sends the text of the email to a Langflow flow.
3. The Langflow flow searches the help documents and writes a draft reply.
4. A Slack node sends the draft to a person for approval.
5. A Gmail node sends the approved reply to the customer.

The HTTP Request node sends the same request as this curl command. The format comes from the Langflow API documentation:

curl -X POST "$LANGFLOW_URL/api/v1/run/$FLOW_ID" \
  -H "Content-Type: application/json" \
  -H "x-api-key: $LANGFLOW_API_KEY" \
  -d '{"input_value": "Text of the support email", "input_type": "chat", "output_type": "chat"}'

Obey these rules when you configure the request:

• Make an API key in Langflow. Langflow 1.13 needs an API key for /api/v1/run by default.
• Keep the API key in an n8n credential, not in the node settings as plain text.
• Send one test request before you map the answer in n8n. The Langflow documentation shows two different paths to the answer text in the JSON response, so read your own response.

n8n publishes a template that calls a Langflow backend from an HTTP Request node. You can also connect the two tools with MCP. The n8n MCP Client Tool can connect to the MCP server of a Langflow project. Neither vendor describes this pair, so test it before you use it in production.

How much do n8n and Langflow cost?#

n8n and Langflow are both free to self-host. You pay for the server, the database, and the calls to the AI model API. n8n also sells n8n Cloud and paid licenses for some self-hosted features. In November 2026, I found no public price or sign-up for a managed Langflow cloud.

On 9 April 2026, DataStax removed hosted Langflow from Astra. Its release note tells users to use Langflow OSS. The IBM Langflow page mentions Langflow Cloud and watsonx Orchestrate, but it shows no price. The Langflow website has a link for a free cloud account, but on 11 November 2026 that link opened the Langflow Desktop page.

The n8n price page showed these n8n Cloud plans in November 2026. The page showed prices in euros for a yearly plan. It also showed that a yearly plan saves 17%.

n8n counts one execution for each run of a full workflow. A workflow with 2 nodes and a workflow with 20 nodes each count as one execution.

The free n8n Community Edition does not have Git version control, environments, external secrets, log streaming, or SSO. It does have queue mode, which is the n8n setup for more volume.

Add the cost of the AI model to the cost of the tool. The RAG development cost guide shows how to calculate the model and server costs. To see if a workflow will save enough time to pay for itself, use the automation ROI calculator.

How do you run n8n and Langflow in production?#

You run both n8n and Langflow in Docker or on a server that you control. The main differences are the scale setup and the next major version of n8n.

Run n8n in production#

The n8n Docker image is n8nio/n8n, and the editor listens on port 5678. n8n uses SQLite by default. For more volume, n8n has queue mode. Queue mode uses a main instance, worker processes that run the executions, and Redis, which holds the queue of jobs. n8n recommends PostgreSQL for queue mode and does not support SQLite in that setup.

The n8n 3.0 beta started on 27 October 2026. n8n expects the stable release in early December 2026. These changes affect a self-hosted server:

• n8n 3.0 needs Docker. n8n no longer supports installs with npm or npx n8n.
• n8n 3.0 removes old nodes, such as Function, Item Lists, Cron, and Interval.
• The default timeout of the Code node falls from 5 minutes to 1 minute.
• n8n 2.x will get security and critical fixes for 12 months after the 3.0 stable release.

Caution: Make a backup of the n8n database before you upgrade to 3.0. The n8n 3.0 notes say that you cannot roll back to 2.x.

Run Langflow in production#

Langflow needs Python 3.10 to 3.14. The documentation recommends uv, which is a Python package manager. Langflow needs 2 GB of RAM at minimum, and the documentation recommends 4 GB. In a Python environment, these two commands install and start Langflow on port 7860:

uv pip install langflow
uv run langflow run

For production, the Langflow best practices separate the editor from a runtime that has no editor. The runtime serves the API only. The documentation recommends an external PostgreSQL database and gives Helm charts for Kubernetes, a system that runs containers on many servers. The documentation gives a minimum of 3 copies of the runtime, each with 2 GiB of RAM and 1 CPU.

To keep a Linux server safe under either tool, read the guide to Linux server hardening with a firewall and Tailscale.

What security risks do n8n and Langflow have?#

Both n8n and Langflow had critical security bugs in 2025 and 2026, and attackers used some of them. Both tools run code and keep the API keys of many systems. Thus one break-in can expose each system that the tool connects to.

A CVE is a public ID for a known security bug. CISA, the US Cybersecurity and Infrastructure Security Agency, lists the CVEs that attackers use in its Known Exploited Vulnerabilities (KEV) catalog.

Both vendors made their defaults safer after these bugs:

• Langflow 1.13 sets LANGFLOW_AUTO_LOGIN=false in its official packages and Docker images. The server does not start without a superuser password, and it refuses the old default password. Most API endpoints and webhooks need an API key. The Langflow authentication documentation gives the details.
• Langflow 1.13.2 lets only admins write custom code by default. The release notes give the details.
• n8n 2.0 enabled task runners by default. A task runner runs the code of the Code node in a separate process. The Code node can no longer read environment variables. The Execute Command node and the Local File Trigger node are off by default, says the list of n8n 2.0 changes.

Warning: Do not put the editor of n8n or Langflow on the public internet. An attacker with access to the editor can read the keys for each connected system.

Do these steps on each server:

1. Update to the latest stable release.
2. Put the editor on a private network or a VPN, such as Tailscale.
3. Make public only the webhook or API paths that outside systems need.
4. In Langflow, keep LANGFLOW_AUTO_LOGIN=false and set a strong superuser password.
5. Give each API key only the permissions that the workflow needs.

Which should you choose: n8n, Langflow, or both?#

Choose n8n for most business automation, and choose Langflow when the AI app is the main product. Use both when a business process has a complex AI step in the middle.

Choose n8n when:

• The workflow starts from an app event, a form, or a schedule.
• The workflow must update a CRM, a help desk, or other business apps.
• A person must approve an action before it occurs.
• People without code skills must read or change the workflow.
• You want a managed cloud service from the vendor.

Choose Langflow when:

• The product is a chatbot, an agent, or a RAG app that other software calls.
• A Python developer will add custom components.
• You need an MIT license, for example to put the tool inside a product that you sell.
• You want to give your flows to other AI apps as MCP tools.

Use both when the process needs many app integrations and the AI step needs fast changes in Python. n8n runs the trigger, the apps, and the approval. Langflow runs the AI step behind one API call.

Use neither when the logic has many branches, the volume is high, or the workflow must have automated tests. In that case, custom code is often a better choice. The guide to n8n vs custom Python for AI workflow automation builds the same workflow both ways.

If you are not sure which tool fits your workflow, describe the workflow for an AI workflow automation quote.

Frequently asked questions#

Is Langflow open source?#

Yes. Langflow uses the MIT license, which lets you use, change, and sell the software. You can self-host Langflow at no cost.

Is n8n open source?#

No. n8n uses the Sustainable Use License, which n8n calls "fair-code". The license lets you use and change n8n for your own internal business purposes or for personal use, but it limits other commercial use.

Can n8n replace Langflow?#

n8n can replace Langflow for simple AI agents and basic RAG steps inside a workflow. Langflow is a better fit when the AI app is the product and developers must extend it in Python. Langflow also gives each flow its own API and MCP tool.

Does Langflow have a cloud version?#

In November 2026, I found no public price or sign-up for a managed Langflow cloud. DataStax removed hosted Langflow from Astra on 9 April 2026. You can self-host Langflow, or use Langflow Desktop on macOS or Windows.

Can Langflow run a flow on a schedule?#

Langflow 1.13 has no documented scheduler. Its code contains a Schedule trigger, but the user documentation does not describe it. Until the documentation describes this trigger, call the run endpoint of the flow from cron or from an n8n Schedule Trigger node.

Sources#

These are the main sources. I checked all sources on 11 November 2026.

• n8n GitHub repository and n8n 2.47.4 release: stars, version, and the license file
• n8n price page: plans, executions, and prices in euros
• n8n Community Edition features
• n8n queue mode
• n8n 2.0 breaking changes and Introducing n8n 3.0 (27 October 2026)
• n8n cluster nodes, MCP Client Tool, and LangChain in n8n
• n8n human-in-the-loop for AI tools and n8n evaluations
• n8n vs. Langflow, a page that n8n wrote about a competitor
• Langflow GitHub repository and Langflow 1.13.2 release: stars, version, and the MIT license
• Langflow installation, production best practices, and API keys and authentication
• Langflow run endpoint (flow trigger endpoints), Webhook component, trigger webhooks, trigger listeners, MCP server, MCP client, custom components, and bundles, with the Slack, Google, and Microsoft triggers
• The Langflow Schedule trigger code in 1.13.2 and pull request 16322
• DataStax Astra release notes: removal of hosted Langflow on 9 April 2026
• IBM to acquire DataStax (25 February 2025) and the IBM Langflow page
• NVD, the CISA KEV catalog, the IBM advisory for CVE-2026-9198, and the GitHub advisory for CVE-2026-25049: CVE details and KEV dates