Langflow
6.9/10Langflow is an open-source, Python-based visual builder that developers use to create, test and deploy AI agents, RAG pipelines and MCP servers as APIs.
Data last updated November 2026 · Our methodology
Overview
Langflow is an MIT-licensed visual framework for building AI applications. Developers drag components onto a canvas and wire them into flows: language models, embedding providers, vector stores, agents and tools. They test each flow in a built-in Playground, then serve it through a REST API or expose it as an MCP server. Any component can be rewritten in Python, and agents can call other components or whole flows as tools. It suits developers and technical teams who want to prototype RAG chatbots and agent workflows quickly while keeping access to the code. It runs as a pip package, a Docker image, or Langflow Desktop for Windows and macOS. The integration library covers major LLM providers, vector databases such as Milvus, Qdrant, Weaviate and MongoDB, and business apps such as Google Workspace, Microsoft 365, Slack, Notion and Confluence. Since version 1.12, the default install includes only a curated set of providers, and the others are installed as separate extension bundles. Logspace created Langflow and DataStax bought it in 2024. IBM bought DataStax in 2025. The software is free. The only paid offering is IBM Elite Support for Langflow, which is quote-only, and IBM also lists a watsonx Orchestrate integration. DataStax removed its hosted Langflow from Astra on April 9, 2026, so most buyers now self-host and pay for their own infrastructure and model APIs. The main drawback is operational: users report instability between versions, slow responses and clutter on large canvases. Self-hosted instances also need prompt patching. Two critical authentication and code-execution flaws affected every version up to 1.12.8 and were fixed in 1.13.0.
Our Rating
Who Langflow is best for
Best software for DevelopersGood fit
It is Python-based and open source, with editable component code, a REST API, MCP server and client support, and pip, Docker and Desktop installs.
There is no license fee, and a basic RAG prototype can be built in minutes, so it suits teams with engineers but a small budget.
An independent reviewer used it for client agent projects and reported much faster delivery than hand-coding.
Less suited
IBM sells SLA-backed support, and version 1.13 adds audit logging and user, team and role management, but the company must self-host and patch it.
IT teams can run it on their own infrastructure with OpenTelemetry monitoring, but they carry the hardening and upgrade work.
Without a managed hosted service and with AI concepts such as embeddings to learn, non-technical small firms will struggle.
Pros & Cons
Pros
- The MIT license means no software fees and full source access. Teams can change any component in Python instead of being limited to the built-in nodes.
- Flows can be served as REST APIs or exposed as MCP servers, so agents built on the canvas plug straight into other apps and MCP clients.
- It includes ready-made RAG building blocks such as document loaders, chunking, embeddings and many vector store connectors. One review found a basic RAG pipeline took about 10 to 15 minutes, against 1 to 2 hours in LangChain code.
- Developers can test each component or flow on its own in the Playground before connecting it, which speeds up iteration.
- It is very popular open-source software, with about 156,000 GitHub stars, and is now backed by IBM, which sells SLA support.
Cons
- DataStax shut down its hosted Langflow on Astra on April 9, 2026, so production use usually means self-hosting and running it yourself.
- Its security record is weak. Two CVSS 9.8 flaws covering authentication bypass and code execution affected versions 1.0.0 to 1.12.8, and several older Langflow flaws are in CISA's Known Exploited Vulnerabilities catalog.
- Users report instability between versions and slow responses, with some flows taking 10 to 30 seconds or more, so pinning versions and testing upgrades is recommended.
- The canvas becomes hard to manage beyond roughly 20 to 30 nodes, and reviewers describe limited multi-agent orchestration.
- Reviewers say the documentation and error messages are thin, which makes debugging tedious.
Key Features
Langflow Pricing
Langflow OSS (self-hosted)
- 100+ integrations
- Community support
- MIT open-source license
- REST API and MCP server
- Python custom components
- Visual flow builder and Playground
- Langflow Desktop for Windows and macOS
- Infrastructure and model API costs paid separately
IBM Elite Support for Langflow
- Price by quote from IBM
- Help with general and technical questions
- SLA-backed help from IBM's Langflow experts
- Paid support subscription for open-source Langflow deployments
Frequently Asked Questions
Yes. The software is open source under the MIT license and costs nothing. You pay separately for hosting, LLM API usage and any vector database. Third-party estimates put a small cloud server at roughly 24 to 33 USD a month.
DataStax removed its hosted Langflow from Astra on April 9, 2026 and pointed users to Langflow OSS. The langflow.org homepage still invites visitors to sign up for a free cloud account, and IBM's product page mentions Langflow Cloud and a watsonx Orchestrate integration. Because these sources conflict, confirm with IBM whether a managed option is currently available.
IBM sells IBM Elite Support for Langflow, a subscription that adds SLA-backed help for open-source deployments. Pricing is not published and comes by quote from IBM.
Logspace, a company founded in January 2022, released Langflow in March 2023. DataStax acquired Logspace in April 2024, and IBM completed its acquisition of DataStax on May 28, 2025.
Version 1.13 requires login by default and adds append-only audit logging, enabled with LANGFLOW_AUDIT_ENABLED. Superusers manage users, teams and roles through the API or the langflow admin command. API keys carry the privileges of the user who created them.
It needs active patching. An IBM bulletin from October 28, 2026 covered critical authentication-bypass and code-execution flaws in versions 1.0.0 to 1.12.8, with 1.13.0 as the only fix. IBM also advises keeping the instance and its MCP endpoint off the public internet.
Options include a pip install, the official Docker image on port 7860, Kubernetes, Langflow Desktop for local development, or running it as a dependency inside your own Python app. Flows run through the API with a Langflow API key.
Since version 1.12, Langflow emits OpenTelemetry traces, metrics and logs to any OTLP-compatible backend. It also integrates with LangSmith and Langfuse for tracing LLM calls.
Similar Tools
See all alternatives
n8n is a workflow automation and AI agent platform that technical teams can run on n8n Cloud or self-host, connecting LLMs to hundreds of business apps.
LangChain offers open-source frameworks (LangChain, LangGraph, Deep Agents) plus the paid LangSmith platform. Developers use them to build, evaluate, deploy and monitor LLM agents.
Open-source platform for building LLM apps, agents, RAG knowledge pipelines and visual AI workflows, available as cloud SaaS or self-hosted.
Ranked near Langflow in AI Software
See Langflow in the listLangflow is #55 in AI Software by our score.
