Langflow
Langflow: visual AI agents, retrieval workflows and Python control
Langflow is a visual builder for AI agents, retrieval applications and MCP services. Connect model, data and tool components, inspect their behavior in the Playground and extend the graph with Python. Developers can run the software locally or deploy their own server, then call flows through an API.
FreeAccount optional
- Access
- Free
- Available on
- Web, API
Connect AI components on a canvas, then extend them with Python
Langflow turns an AI application into a graph of components. Prompts, models, data sources and tools are connected visually, so their inputs and outputs are visible while the application is being built. The Playground provides a conversational interface for trying a flow and inspecting its responses. Python customization adds behavior beyond the prebuilt components.
The Agent component combines model-provider configuration, instructions and tool calling. Components can become tools, individual tool actions can be enabled or disabled, and agents can use other agents or flows. Provider credentials determine which models are available and which account pays for calls. Retrieval applications can combine data sources, embeddings and vector stores alongside their model components.
Human review can be added through a Human Input component or selected tool approvals. A gated tool call pauses at a checkpoint until a decision is made in the Playground; completed steps do not need to run again. A nested Run Flow target cannot contain its own approval pause, so that review belongs in the parent process.
Developers can run Langflow locally or deploy a server using containers, remote infrastructure or Kubernetes, with APIs exposing the flows to other applications. The MIT licence permits reuse subject to its notice requirements. The editor can execute Python with access to its host resources, and a shared process does not provide tenant isolation. Production authentication, isolation and connected-service data handling depend on the deployment.
Best for
- Developers building AI agents and retrieval applications
- Teams extending visual flows with Python
- Self-hosted AI services with API and MCP integration
Limitations
Free software does not include paid model calls, databases or hosting. Production deployment requires technical ownership. The Python editor has access to host resources; tenant isolation must come from infrastructure. Approvals are configured per flow or tool, and nested Run Flow targets cannot pause for their own human decision.
Langflow agent and application features
Visual component graph
Connect prompts, models, data sources and tools with visible inputs and outputs.
Agent tool actions
Attach tools to an agent and control which actions are available.
Python customization
Extend the builder with custom components and reusable tool behavior.
Playground and human decisions
Try a flow in chat and resolve configured approval gates from a checkpoint.
Flow and management APIs
Run deployed flows and manage application resources through API endpoints.
Local and server deployment
Use your own runtime and infrastructure, including container or Kubernetes deployment.
Langflow pricing
Free to use. Plans and included features vary. Visit Langflow for full plan details and current offers.
Technical specifications
Automation and agents
| Feature | Langflow |
|---|---|
| Connectors and tool access | Model providers, data sources, vector stores, tools and MCP components on a visual graph. Python custom components can expose additional tool actions.Configured components and provider credentials |
| Triggers and scheduling | Run flows from the editor/Playground or call a deployed flow through the API. Connected components and external callers determine when a process starts.Local or deployed server |
| Human approval controls | A Human Input component or selected agent tools marked Requires approval pause a flow at a checkpoint. Playground decisions approve or reject; nested Run Flow targets cannot themselves pause for HITL.Configured gates; nested-flow approval belongs in parent |
| Hosting and deployment | Run locally or self-host with containers, remote servers or Kubernetes. The IDE executes Python; infrastructure must provide tenant isolation and production access controls.Owner-managed deployment and connected model services |
| Execution accounting | MIT software has no subscription charge. Model APIs, vector stores and infrastructure are independently funded; there is no bundled provider-credit or workflow-execution allowance.Software licence separate from running costs |
| API access | SupportedFlow and management APIs on your Langflow deployment |
Langflow alternatives
Other platforms offer different combinations of visual AI building, hosted operation and app automation.
Dify
Visual AI applications with managed Cloud plans, knowledge bases and Community deployment.
Compare with LangflowMake
Hosted automation with connected apps, branching and AI-agent modules.
Compare with LangflowLangflow FAQs
Is Langflow free?
The software is free under the MIT licence. Model-provider calls, databases and infrastructure can still incur charges through the services you connect.
Can I self-host Langflow?
Yes. Its documentation covers local servers, containers, remote infrastructure and Kubernetes. Production access controls and tenant isolation depend on your infrastructure.
Does Langflow require coding?
The visual editor provides prebuilt components for connecting an application. Python is available for custom components, and deploying or integrating a production service requires technical work.
Can Langflow require approval before a tool runs?
Selected agent tools can require approval, and Human Input components can pause flows. The Playground supports approve/reject decisions; approval steps must sit in the parent when using nested Run Flow.
Is this your product? Claim this page to update your listing.