Helm-Wrapping Dify’s Multi-Service LLM Stack for K8s

What it does

This is a packaging project—a Helm chart that deploys the upstream Dify platform, an open-source LLM chatbot and app builder, onto Kubernetes clusters. It templates the upstream container images (API server, web frontend, Celery workers, plugin daemon, code sandbox, and SSRF proxy) into a coordinated release, handling service wiring, storage, and traffic routing. The goal is to collapse Dify’s sprawling architecture into something you can install and iterate on a standard K8s cluster.

The interesting bit

The chart stands out for its unusually broad menu of supported external backends: you can swap in PostgreSQL or MySQL, Redis Standalone or Sentinel, eight vector databases (including Weaviate, Qdrant, Milvus, and PGVector), and seven object-storage providers ranging from S3 and Azure Blob to Tencent COS and Volcengine TOS. The README also includes a refreshingly explicit Mermaid diagram showing exactly how nginx routes traffic to the API, web, and plugin services.

Key highlights

• Packages the full Dify stack (API, Web, Worker, Beat, Plugin Daemon, Sandbox, SSRF Proxy) as a single Helm release based on upstream 1.12.1 images
• Supports externalizing nearly every dependency: databases, cache, vector stores, and object storage
• Bundles an nginx reverse proxy with predefined routing rules for /api, /console, /files, /e/, and other endpoints
• Targets Kubernetes 1.23+ and Helm 3.12+
• Published to Artifact Hub with automated release workflows

Verdict

Worth exploring if you already run Kubernetes and want to self-host Dify without maintaining dozens of raw manifests. Not for you if you prefer managed hosting or a simple single-node Docker Compose setup.