Custom GPTs, but you actually own the infrastructure

What it does RAGapp wraps LlamaIndex into a single Docker container with an admin UI, chat interface, and auto-generated API. Point it at OpenAI, Gemini, or a local Ollama model, configure your knowledge sources, and you have a deployable RAG agent without writing orchestration code.

The interesting bit The project explicitly offloads the hard parts it doesn’t want to solve: authentication is “by design” absent, punted to whatever API gateway you already have. That’s either refreshing honesty or a deployment headache, depending on your existing stack. The “multiple RAGapps” Compose setup suggests they know enterprise means many teams, not one demo.

Key highlights

• One-liner Docker run; admin UI at /admin, chat at root, OpenAPI docs at /docs
• Supports hosted models (OpenAI, Gemini) and local via Ollama
• Pre-built Compose variants: single-node with Ollama+Qdrant, or multi-tenant with management UI
• Frontends dynamically pulled from create-llama; make build-frontends required before dev commits
• Kubernetes descriptors listed as “coming soon”

Caveats

• No built-in auth; you bring your own API gateway or use their multi-app Compose template
• Authorization via forwarded access tokens is “later versions” territory, not implemented
• Development setup requires Poetry and a Makefile step that fetches upstream frontend code

Verdict Worth a spin if your org wants GPT-style RAG agents behind its own firewall and already has gateway/auth infrastructure. Skip it if you need turnkey multi-user security or are allergic to Docker operations.