langflow-ai/langflow-helm-charts: The Compiler for Visual AI

Moving beyond the canvas to a hardened, headless production runtime for agentic workflows.

• View on GitHub • More from langflow-ai

A massive, ornate Victorian birdcage containing a very small mechanical bird, being hoisted by a heavy crane. This illustrates the bloated payload of deploying a full visual IDE to production just to run a single workflow. — Deploying a visual canvas to production often means hauling heavy frontends and databases just to execute a single prompt.

Key Takeaways

Langflow separates the visual design environment from the execution engine to eliminate resource bloat in production.
The runtime chart transforms static JSON flow files into headless API endpoints without the need for a frontend.
Custom naming overrides prevent Kubernetes environment variable collisions that often break application logic in strict environments.
A dual-chart strategy allows teams to scale stateless AI workloads independently from the stateful prototyping IDE.

The Production Trap

The visual AI trend usually ends at the browser tab. You build a beautiful graph, but then you are stuck running a heavy, fragile IDE in production just to execute a single prompt. The hidden cost of visual AI is the resource bloat and security surface area of running a design tool in a Kubernetes cluster.

I’m encountering an error when deploying a Langflow instance on Kubernetes starting from version 1.7.1 and above. The same setup works without any issues on v1.6.9. The failure happens during the initial setup/startup phase, not at runtime.

vogel-alex, User · Issue with Kubernetes setup · Issue #11445 · langflow-ai/langflow

Deploying the full application leads to inevitable friction. The story here is not just Langflow on Kubernetes. It is the architectural split between the IDE and the runtime. This repository represents the compilation of visual spaghetti into hardened, headless infrastructure.

The Runtime Surgery

The repository provides two distinct Helm charts. The langflow-ide chart deploys the full canvas experience using a StatefulSet for the backend and a Deployment for the React frontend. The langflow-runtime chart takes a completely different approach.

The runtime chart strips away the frontend entirely. It uses an init-style logic block within the main container to download JSON flow files and inject them directly into a Python execution engine. This transforms a generic container into a highly specific API endpoint.

How the runtime chart hydrates a static JSON file into a live, headless API endpoint.

Enterprise Friction and the OpenShift Fix

Standard Kubernetes naming conventions often break in strict environments like OpenShift. If a service is named exactly 'langflow', Kubernetes automatically injects environment variables like LANGFLOW_SERVICE_HOST into the pods.

This creates a collision with the application's internal configuration logic. The Helm charts solve this by heavily utilizing the fullnameOverride helper to rename services, ensuring the application configuration remains untainted by the orchestrator.

Scaling the Frozen Graph

By separating the design environment from the execution environment, teams can scale their AI workloads efficiently. The runtime can be deployed as a standard stateless Deployment, while the IDE retains its StatefulSet to manage SQLite databases during prototyping.

Feature	Langflow IDE Chart	Langflow Runtime Chart
Workload Type	StatefulSet + Deployment	Deployment
Primary Function	Visual canvas and testing	Headless API execution
Storage	Persistent (SQLite/Postgres)	Stateless (JSON injection)
Attack Surface	High (UI + API exposed)	Minimal (API only)

This dual-chart strategy is what elevates Langflow from a local prototyping tool to a production-grade LLMOps platform. It allows developers to build visually, but deploy headlessly.