# Langflow — Open-source visual framework for building AI agents and RAG workflows in Python.

> Source: The Agents Index — https://theagentsindex.com/langflow (structured, researched, re-verified)
> Facts last verified: 2026-09-03

Langflow is an open-source, Python-based framework for building AI agents and retrieval applications by wiring components together in a visual editor. Flows built in it become services you host and serve yourself, and every component can be extended with custom Python.

| Fact | Value |
| --- | --- |
| Website | https://www.langflow.org |
| Free tier | Yes |

The whole product is free open-source software with no paid tier: Docker and pip installs require no account or payment, and no feature is metered. The macOS desktop build is the one gated route, requiring a contact form before download. Inference is not included, so you pay your own model provider.

## What it does

**Writes code** · Yes — Langflow Assistant turns a natural-language prompt into whole flows or into Python for a single custom component, scoped to Langflow artifacts rather than an application codebase.

> Langflow Assistant understands the structure of the Langflow graph, so it can build complete flows or create individual components from natural language prompts.
> — https://docs.langflow.org/langflow-assistant · Langflow Assistant, intro, third paragraph

**Whole-codebase work** · No — The vendor bounds the assistant's context to the single flow open on the canvas, so it cannot carry a change across a repository or even across flows.

> The language model in Langflow Assistant only has the currently opened flow in your workspace for context.
> — https://docs.langflow.org/langflow-assistant · Langflow Assistant, paragraph beginning 'Behind the scenes, Langflow Assistant runs a built-in Langflow flow'

_Not established: Debugs and fixes_

## How much it does unattended

**Runs autonomously** · Yes — Agents you assemble call tools in a loop and decide themselves which to invoke, so a run can finish without a person intervening; approval gates are opt-in.

> Agents extend Large Language Models (LLMs) by integrating tools, which are functions that provide additional context and enable autonomous task execution.
> — https://docs.langflow.org/agents · How do agents work?, first paragraph

**Multi-agent** · Yes — One agent can be put in Tool Mode and attached to another, and the A2A component reaches agents outside the instance, so several agents can share a task.

> To create multi-agent flows, you can set another Agent component to Tool Mode, and then attach that agent as a tool for your primary Agent component.
> — https://docs.langflow.org/agents-tools · Use an agent as a tool, first sentence

**Agent permissions** · Partly — Default is ungated tool calling; approval is opt-in per tool, and then the run checkpoints and waits for a human Approve or Reject before continuing.

> The Agent component can pause before calling a gated tool, create a stateful checkpoint, and then resume after a human decision.
> — https://docs.langflow.org/agents-tools · Require approval for agent tools, first paragraph

## Which models it runs on

**Claude** · Yes — An Anthropic bundle supplies Claude chat models, and the vendor notes basic Anthropic support ships in every install through the core Language Model component.

> The Anthropic component generates text using Anthropic Chat and Language models like Claude.
> — https://docs.langflow.org/bundles-anthropic · Anthropic text generation, first paragraph

**GPT** · Yes — OpenAI is a first-class provider with its own bundle, and the GPT model names are enumerated in the component's own parameter table.

> Input parameter. The name of the OpenAI model to use. Options include "gpt-3.5-turbo" and "gpt-4".
> — https://docs.langflow.org/bundles-openai · OpenAI text generation parameters table, model row

**Open models** · Yes — Open-weight models run through local servers: Ollama, vLLM and LM Studio each have components that point at an instance you operate.

> To use the Ollama component in a flow, connect Langflow to your locally running Ollama server and select a model:
> — https://docs.langflow.org/bundles-ollama · Ollama text generation, instructions intro

**Model choice** · Yes — You choose the provider and model per component; the vendor states plainly that no particular model or vector store is required.

> It supports important AI functionality like agents and the Model Context Protocol (MCP), and it doesn't require you to use specific large language models (LLMs) or vector stores.
> — https://docs.langflow.org/ · What is Langflow?, opening paragraph

## Where you use it

**In your editor** · No — The vendor enumerates every way to run Langflow and none is an editor plugin: it is a desktop app, a container, or a Python package serving a browser canvas. IDEs reach it only as MCP clients.

> Docker Run Langflow in a container in isolation. Python package Full control over environment, dependencies, and versioning.
> — https://docs.langflow.org/get-started-installation · Install Langflow, the card list enumerating the install options (Desktop, Docker, Python package, install from source)

**On the command line** · Yes — Two terminal surfaces exist: the langflow CLI that runs and manages the server, and lfx for one-shot flow runs and serving.

> The Langflow command line interface is the main interface for managing and running the Langflow server.
> — https://docs.langflow.org/configuration-cli · Langflow CLI, opening line

**In your pipeline** · Partly — The DevOps toolkit scaffolds GitHub Actions to validate, test and push flows, but nothing claims Langflow itself reviews or acts on a pull request.

> .github/workflows: GitHub CI tooling.
> — https://docs.langflow.org/flow-devops-sdk · Create a project and version a flow, step 3 project scaffold list

**In a browser** · Yes — The visual editor and Playground are served in a browser by the instance you start; there is no vendor-run SaaS in the deployment docs.

> To confirm that a local Langflow instance is running, navigate to the default Langflow URL http://127.0.0.1:7860.
> — https://docs.langflow.org/get-started-installation · Install and run the Langflow OSS Python package, step 5

## Whose machine it runs on

**Self-hosted** · Yes — Self-hosting is the only hosting the docs describe: a published Docker image, a Helm chart for production Kubernetes, or the Python package.

> You can use the Langflow Docker image to start a Langflow container.
> — https://docs.langflow.org/get-started-installation · Install and run Langflow with Docker, step 2 intro

**Open source** · Yes — The vendor calls its own codebase open source and documents installing from the public repository; the docs themselves do not name the licence in prose.

> As an open-source codebase in a rapidly developing field, Langflow welcomes contributions, whether it be in the form of a new feature, improved infrastructure, or better documentation.
> — https://docs.langflow.org/contributing-how-to-contribute · Contribute to Langflow, intro, second sentence

## What it costs to run

**How it meters** · No — Langflow meters nothing at all: no seats, credits, tokens or runs. Inference cost lands on the provider account whose key you paste in.

> The key must have permission to call the models you want to use in your flow, and your account must have sufficient credits for the actions you want to perform.
> — https://docs.langflow.org/get-started-quickstart · Run the Simple Agent template flow, step 2 global model provider configuration

**Free tier** · Yes — Docker and pip installs need no account, key or payment step; only the desktop build is gated behind a contact form.

> Langflow can be installed in multiple ways. Choose an option below to jump to the steps.
> — https://docs.langflow.org/get-started-installation · Install Langflow, intro sentence

**API access** · Yes — Every flow is callable over HTTP with a Langflow API key, and the vendor ships a TypeScript client plus generated curl, Python and JavaScript snippets.

> The Langflow API is the primary way to access your flows and Langflow servers programmatically.
> — https://docs.langflow.org/concepts-publish · Use the Langflow API to run flows, first sentence

**MCP server** · Yes — Each project runs its own MCP server so an external agent can call your flows as tools, and Langflow is an MCP client in the other direction.

> Langflow integrates with the Model Context Protocol (MCP) as both an MCP server and an MCP client.
> — https://docs.langflow.org/mcp-client · Page intro, first sentence

**Bring your own key** · Yes — You paste your own provider keys into the Model Providers pane, and the OpenAI-compatible provider points at any endpoint you host yourself.

> In the API Key field, add your provider's API key. Some providers require additional configuration fields.
> — https://docs.langflow.org/components-models · Use language model components in flows, step 1.3 Model Providers pane

## Buying it for a team

**A company can buy it** · Partly — No software licence is sold, so there is nothing to procure for the engine itself; an organisation can only buy an IBM support subscription around its own deployment.

> IBM Elite Support for Langflow is a subscription to the Langflow expertise at IBM.
> — https://docs.langflow.org/luna-for-langflow · IBM Elite Support for Langflow, paragraph 2

**Seat model** · No — Multi-user accounts and roles exist, but no seat price, seat floor or seat ceiling is published anywhere: the software is free to run for as many users as you provision.

> Langflow seeds three built-in system roles that an enforcement plugin uses as a starting point for policy sync. These roles cannot be deleted.
> — https://docs.langflow.org/authorization · System roles, intro sentence above the role table

**Admin controls** · Yes — A superuser can hide components, templates, providers and models from the builder, block custom components, and restrict which fields API callers may override.

> Superusers can create catalog and model policy to hide components, templates, model providers, and models from the Langflow visual builder.
> — https://docs.langflow.org/api-governance-policy · Catalog and model policy, second paragraph

**Audit log** · Partly — An admin audit table exists but is off by default, covers only authorization decisions and share administration, and the vendor calls the default producer best effort.

> Write an audit log row for every authorization decision and share-administration action.
> — https://docs.langflow.org/authorization · Configure authorization environment variables table, LANGFLOW_AUTHZ_AUDIT_ENABLED row

**Single sign-on** · Partly — OIDC works by validating a token an upstream proxy forwards, with just-in-time user creation from claims; SAML is not mentioned anywhere and Langflow does not run the login itself.

> External authentication lets an upstream identity provider, OIDC proxy, or corporate SSO gateway handle login.
> — https://docs.langflow.org/external-authentication · External authentication, first paragraph

## What happens to your code

**Opt out of training** · Partly — Anonymous usage telemetry is on by default and switched off with one variable, but the vendor makes no statement about training models on your data either way.

> To opt out of telemetry, set DO_NOT_TRACK=True in your Langflow environment variables before starting Langflow. This disables telemetry data collection.
> — https://docs.langflow.org/contributing-telemetry · Opt out of telemetry, first sentence

**Data residency** · Yes — Jurisdiction is whichever one you deploy into: flows, message history and logs sit in a SQLite file or a Postgres instance you operate.

> In this configuration, all structured application data from Langflow, including flows, message history, and logs, is instead managed by PostgreSQL.
> — https://docs.langflow.org/configuration-custom-database · Paragraph following the SQLite callout

**Getting out** · Yes — Flows export as plain JSON from the UI or the API, individually or as a zipped project, and there are no credits that could expire.

> Exported flows are downloaded to your local machine as JSON files named FLOW_NAME.json. If you export an entire project, the JSON files are packaged in a zip archive.
> — https://docs.langflow.org/concepts-flows-import · Export a flow, paragraph after the three export methods

**Certifications** · No — No SOC 2, ISO 27001 or GDPR attestation is claimed; the single GDPR mention assigns the obligation to whoever operates the server.

> Follow industry best practices and legal requirements for data privacy and encryption of data in transit and at rest, including GDPR requirements, HTTPS, TLS, and SSL.
> — https://docs.langflow.org/deployment-prod-best-practices · Security, Encryption and privacy bullet

## Provenance

Every fact above comes from the vendor. Nothing here is independently corroborated yet.

## Coverage elsewhere

None recorded. Every fact above comes from the vendor and is not independently corroborated yet.

## What changed

- 2026-09-03 — Capabilities · Commercial terms · Free tier

## Correcting this record

Anything marked "Not established" is a question we have not answered — it is not a No. If you represent Langflow, or anything here is wrong or out of date, you can supply it with a source: the operations this site accepts from an agent are listed at https://theagentsindex.com/.well-known/agents.json — that catalogue needs no key to read, and each entry states its own auth.

Supplying evidence is free. Nothing that can be bought here changes what a field says, whether a gap is closed, or how this record is scored.