AI agents sound powerful, but wiring one up usually means code. Langflow removes that barrier: it’s an open-source platform for building and deploying AI-powered agents and workflows through a visual, drag-and-drop interface — with Python access when you want it. This tutorial shows you how to install Langflow, build your first agent visually, and deploy it as an API your business can actually use.
What It Is
Langflow is a visual authoring platform for AI agents and workflows. You drag components — language models, prompts, vector stores, tools — onto a canvas, connect them into flows, test in an interactive playground, and then deploy each flow as an API or MCP server that any application can call. As checked on October 18, 2026, it has 155,785 stars on GitHub (github.com/langflow-ai/langflow) and is licensed under the MIT License.
Highlight features from the README: a visual builder for fast iteration, source code access to customize any component in Python, an interactive playground with step-by-step control, multi-agent orchestration with conversation management and retrieval, deploy as an API or export flows as JSON for Python apps, deploy as an MCP server so flows become tools for MCP clients, observability integrations (LangSmith, LangFuse), and enterprise-ready security. It supports all major LLMs and vector databases, and ships with a growing library of AI components. There is also a Langflow Desktop app for Windows and macOS that bundles all dependencies.
A note on Flowise: you may see the older project FlowiseAI/Flowise recommended in old guides, but that repository has been archived by its maintainers. Langflow is the active, maintained equivalent for visual agent building — this tutorial uses Langflow.
Who It’s For / SMB Use Case
Langflow fits small businesses that want custom AI behavior without hiring an AI engineering team. Practical uses include:
- Support agent with your docs: Build a flow that retrieves answers from your knowledge base and escalates to a human when confidence is low.
- Lead research agent: A flow that takes a company name, searches the web, and returns a structured brief for your sales team.
- Document processing pipeline: Ingest invoices or applications, extract fields with an LLM, and output clean JSON to your systems.
- Internal tools as APIs: Deploy a flow as an API endpoint and call it from your website, CRM, or n8n workflows.
Langflow is not a finished chatbot product — it’s a builder. If you want an out-of-the-box document chatbot with no building at all, a product like AnythingLLM is simpler. Langflow pays off when you need custom logic, multiple tools, or an API you control.
Installation
The README recommends installing locally with Python 3.10–3.14 and the uv package manager. There is also a Desktop app for Windows and macOS, and a Docker option.
Local install (recommended)
From a fresh directory, install and run:
uv pip install langflow -U
Then start Langflow:
uv run langflow run
Open your browser at http://127.0.0.1:7860. The README notes that official release packages disable automatic login by default: set the LANGFLOW_SUPERUSER_PASSWORD environment variable before starting, then sign in with the username langflow.
Docker install
Set a strong admin password and run the official image:
read -rsp "Langflow administrator password: " LANGFLOW_SUPERUSER_PASSWORD; echo export LANGFLOW_SUPERUSER_PASSWORD docker run -p 7860:7860 -e LANGFLOW_SUPERUSER_PASSWORD langflowai/langflow:latest
Langflow is then available at http://localhost:7860; sign in as langflow with your password.
Desktop app
For the simplest start on Windows or macOS, download Langflow Desktop from langflow.org — all dependencies are included, so you skip Python environment setup entirely.
How to Use Langflow
Step 1: Start from a template
On the Langflow home screen, browse the template gallery and pick something close to your goal — for example, a “Simple Agent” or “RAG” template. Templates are pre-wired flows you can inspect and modify, which is far faster than starting from a blank canvas. Open it and you’ll see components connected by edges on the canvas.
Step 2: Configure your model component
Click the language model component in the flow and enter your API key (OpenAI, Anthropic, Google, or another provider). Choose the model and adjust temperature: lower for factual tasks, higher for creative ones. If you prefer local models, Langflow also supports Ollama-style local endpoints.
Step 3: Add tools and knowledge
Drag in the components your agent needs: a vector store component connected to your documents for RAG, a web search tool for fresh information, a calculator for math, or custom API tools. Connect them to the agent component’s tool inputs. Keep the toolset small at first — every tool you add is another decision the agent must make.
Step 4: Test in the playground
Open the Playground and run real queries through your flow. Langflow shows step-by-step execution, so you can see which components ran, what the model received, and where things went wrong. Iterate on the prompt text and component settings until the behavior is reliable on your test cases.
Step 5: Deploy as an API or MCP server
When the flow works, use the API tab to get an endpoint you can call from your website, app, or automation platform — or deploy it as an MCP server so MCP-compatible clients can use your flow as a tool. You can also export the flow as JSON to version-control it or move it between environments. Set up observability with LangSmith or LangFuse to monitor production behavior.
Pricing / Limitations
Langflow is free and open source under the MIT License — no seat fees, no feature gates. Your costs are hosting (your own machine, a VPS, or a cloud deployment) and LLM API usage, which scales with how much your flows are called. Deployment guides on docs.langflow.org cover all major clouds.
Limitations: you still need to understand agent concepts — prompts, tools, retrieval — to build something reliable; the visual interface removes coding, not thinking. Complex multi-agent flows can be slow and expensive at runtime, so test costs with real volumes before launching. Local install requires managing a Python environment (the Desktop app or Docker avoids this). And as with any agent framework, keep humans in the loop for irreversible actions and monitor outputs in production.
Bottom Line for SMB
Langflow is the fastest way for a non-specialist team to go from “we need a custom AI agent” to a working, deployable API — visually, in days not months. Start with one well-defined flow (a support agent over your docs is the classic first win), prove it in the playground, then deploy and monitor. For background on agent concepts, read our AI agents explained guide, and if you want to build agents in code later, our LangChain tutorial is the natural next step.
Frequently Asked Questions
Do I need to know Python to use Langflow?
No. Flows are built visually by connecting components, and templates give you working starting points. Python is optional — available for customizing components when you outgrow the visual options.
What’s the difference between Langflow and the archived Flowise project?
Both are visual builders for AI agents, but FlowiseAI/Flowise has been archived by its maintainers and is no longer developed. Langflow is actively maintained, MIT-licensed, and adds features like MCP server deployment and multi-agent orchestration — use Langflow for any new project.
How do I put a Langflow agent on my website?
Deploy the flow and use the generated API endpoint: your website’s backend calls it like any REST API and displays the response. Alternatively, deploy the flow as an MCP server and connect it to MCP-compatible clients. Check the deployment guides on docs.langflow.org for your hosting platform.
