Langflow

langflow-ai/langflowReviewed by AI

What it does

Langflow is a platform where you draw an AI workflow instead of writing it: you drag boxes for models, databases, files and tools onto a canvas, join them, and try the result in a built-in playground. Each finished flow can then be served as an API or as an MCP tool for other programs. It is aimed at developers, but the Desktop version for Windows and macOS includes everything needed, so no Python setup is required. You still have to connect your own AI service, and most of those charge for use.

How easy it is to install

B · installer plus one simple step · Windows

Windows B
macOS B
Linux C

On Windows and macOS there is Langflow Desktop, and the README says "All dependencies are included, so you don't need to manage Python environments or install packages manually" — after installing you still have to paste a key for an AI service. On Linux there is no desktop build: you install Python 3.10–3.14 and the `uv` tool, run `uv pip install langflow -U`, set a `LANGFLOW_SUPERUSER_PASSWORD` and then `uv run langflow run` before opening http://127.0.0.1:7860.

Numbers on GitHub

Stars: 156,203on Oct 31, 2026
Downloads: 225,9451,628 this week
Growth: +161stars in 5 days

Stars over time

Facts

Runs on: Windows, macOS, Linux
You need to install: Nothing extra
To build from source: Python, Node.js, Graphics card
Disk space: 335 MB–1,005 MB · from the installer
Accounts or keys: OpenAI, Anthropic, Langflow (its own key, for calling your flows from other programs)
Cost: Free, API costs extra · The project is free and MIT-licensed. Flows only do something once you connect an AI service (OpenAI, Anthropic, Google and many others), and those bill you for what you use; a model running locally through Ollama is the free route.
License: MIT (Permissive)
Can replace: Zapier, Make
Last update: Oct 30, 2026

Risk signals (5)

Attentionruns python code without a sandbox by default
The README offers "Source code access lets you customize any component using Python", and pyproject.toml describes the isolation as optional: "Opt-in microVM sandbox backend for code-execution components (LANGFLOW_SANDBOX_BACKEND=exec-sandbox, issue #12029). Requires Python >= 3.12 and QEMU 8+ on the host; on older Pythons the extra installs nothing". A flow you import from someone else can therefore run code on your machine unless you switch that on.
AI review
Attentionssrf protection off by default
.env.example: "SSRF (Server-Side Request Forgery) Protection — When enabled, blocks requests to private IP ranges, localhost, and cloud metadata endpoints ... Default: false (for backward compatibility)". Left as shipped, a flow can be pointed at addresses inside your own network.
AI review
Infoflows can be run without signing in
.env.example documents "LANGFLOW_PUBLIC_FLOW_RATE_LIMIT_PER_MINUTE: Number of anonymous public-flow runs allowed per minute per IP" with a default of 20, so a flow published as public can be run by anyone who reaches the server, spending your API credits.
AI review
Infologin required in official builds
README: "Official release packages and tagged source archives disable automatic login by default. Set `LANGFLOW_SUPERUSER_PASSWORD` before starting", while "Development branches retain automatic login by default". If you run it from a development branch there is no password until you set one.
AI review
Infovery large dependency list
pyproject.toml pulls in 16 provider bundles by default (lfx-ibm, lfx-openai, lfx-anthropic, lfx-google, lfx-slack and more) and carries a block of version floors added specifically to dodge published vulnerabilities ("oauthlib>=4.0.0,<5.0.0 # CVE-2026-49264, CVE-2026-49265", "fsspec>=2026.6.0 # CVE-2026-104851", and others). The checks estimate 335–1005 MB of disk.
AI review

Evidence is shown in its original language.

What Radar could not check

Whether the Desktop app installs and runs on your version of Windows or macOS, since Radar did not run it
What a flow downloaded from someone else does when you open and run it
The real monthly cost, which depends on the AI service, vector database and search providers you connect
Real disk use after the optional document, audio and local-model extras are installed
Whether your installation ends up reachable from outside your machine, which depends on how you start it and on your network