🐳 Container Diet

AI-powered Docker image optimization — because your containers could use a diet.

Container Diet analyzes your Docker images and Dockerfiles, then serves up sassy, actionable advice backed by your choice of 12+ AI providers. Catch bloated layers, security holes, and missing best practices before they reach production. Works as a CLI, in CI/CD, or as an MCP tool inside your AI editor.

⚡ Quick Start

# 1. Install
go install github.com/k1lgor/container-diet/cmd/cli@latest

# 2. Configure your AI provider
container-diet init-config
# → edit ~/.config/container-diet/config.yaml
# → uncomment your provider, paste your API key

# 3. Analyze
container-diet analyze --dockerfile Dockerfile --auto-fix

That's it. You'll get a roast of your Dockerfile, actionable fixes, and a Dockerfile.diet with the optimizations applied.

🧠 AI Providers

Container Diet talks to any of these — bring the key you already have, or run locally for free.

Provider	Config Key	API Key Env Var
OpenAI	`openai`	`OPENAI_API_KEY`
Anthropic (native)	`anthropic`	`ANTHROPIC_API_KEY`
OpenRouter	`openrouter`	`OPENROUTER_API_KEY`
Groq	`groq`	`GROQ_API_KEY`
DeepSeek	`deepseek`	`DEEPSEEK_API_KEY`
Mistral	`mistral`	`MISTRAL_API_KEY`
xAI (Grok)	`xai`	`XAI_API_KEY`
Ollama (local, free)	`ollama`	none needed
Perplexity	`perplexity`	`PERPLEXITY_API_KEY`
Moonshot	`moonshot`	`MOONSHOT_API_KEY`
Hugging Face	`huggingface`	`HF_API_TOKEN`
Custom endpoint	`custom`	`CUSTOM_API_KEY`

Switch providers with --provider or set default_provider in your config.

🔌 MCP Server — AI Agent Integration

Container Diet ships as an MCP (Model Context Protocol) server so AI agents can analyze containers directly from your editor.

Supported clients: Claude Desktop · Cursor · Claude Code · Codex · any MCP-compatible agent

Add to your MCP client config:

{
  "mcpServers": {
    "container-diet": {
      "command": "container-diet",
      "args": ["mcp", "server"]
    }
  }
}

No API keys in the MCP config — the server reads them from ~/.config/container-diet/config.yaml.

Tools exposed to AI agents:

Tool	Description
`analyze_dockerfile`	AI analysis with optional auto-fix generation
`analyze_image`	Layer breakdown + AI optimization advice
`get_optimization_advice`	General container advice from free-form context
`get_image_summary`	Quick metrics without burning AI tokens

Full setup guide: container-diet mcp init

📖 Usage

Analyze a Dockerfile

container-diet analyze --dockerfile Dockerfile

Analyze + Auto-Fix

container-diet analyze --dockerfile Dockerfile --auto-fix
# Writes Dockerfile.diet with optimizations applied

Analyze a Docker Image

# Local daemon
container-diet analyze nginx:latest

# Pull from registry
container-diet analyze python:3.12-slim --remote

# Auto-pull if missing locally
container-diet analyze busybox --pull-missing

JSON Output (CI/CD)

container-diet analyze --dockerfile Dockerfile --format json

{
  "advice": "⚠ WARNING: Root user detected...\n✓ SUGGESTION: Use non-root user...",
  "fix": "FROM nginx:alpine\nUSER nginx\n..."
}

Choose Provider + Model

container-diet analyze --dockerfile Dockerfile --provider anthropic --model claude-sonnet-4-6

🎮 Demo

Running the "Nightmare Monolith" Dockerfile through Container Diet:

📄 Reading Dockerfile: samples/Dockerfile.nightmare...

🤖 [AI ANALYSIS]
🚢 Asking the Container Dietician for insights using openrouter (openai/gpt-4o-mini)

================================================================

Oh, honey, what do we have here? A "Monolith Monster" Dockerfile that's
about to sink your ship with its fatty layers and spicy security risks! 🚀

---

⚠ WARNING: Version Drift Alert!
Using `ubuntu:latest` means you're playing roulette with your build
environment. 🎰

✓ SUGGESTION: Pin your base image to a specific version like
`ubuntu:22.04` to keep things predictable.

---

⚠ WARNING: Apt-get Avalanche!
Installing everything but the kitchen sink? This is the definition of
bloat, my dear. 🐘

✓ SUGGESTION: Ditch `openjdk-11-jdk`, `build-essential`, `cmake`, and
`gdb` unless you truly need them.

---

⚠ WARNING: 777 permissions? Root SSH? Root login permitted?
You might as well hand over the keys to the universe. 🔑

✓ SUGGESTION: Use non-root user, restrictive permissions. And ask
yourself — do you really need SSH in a container?

---

Oh, darling, trim that bloated ship before it swallows a sun! 🐳✨

⚙️ Configuration

Config file locations (searched in order, later overrides earlier):

Priority	Path	Scope
1 (highest)	`--config <path>` flag	Per-command
2	`.container-diet/config.yaml`	Per-project
3 (lowest)	`~/.config/container-diet/config.yaml`	Global

Minimal config example:

default_provider: "openrouter"

providers:
  openrouter:
    api_key: "${OPENROUTER_API_KEY}" # or paste the key directly
    base_url: "https://openrouter.ai/api/v1"
    default_model: "openai/gpt-4o-mini"
    timeout_seconds: 90

analysis:
  max_tokens: 4096

ui:
  theme: "docker"
  show_emojis: true

Generate the full example with all providers: container-diet init-config

Pro-tip: use --local for project-specific config, --force to overwrite.

🏗️ Architecture

cmd/cli/main.go               # Entry point
internal/
├── ai/                       # Provider interface + implementations
│   ├── provider.go           # Interface, prompt builders, retry logic
│   ├── anthropic.go          # Anthropic native API
│   └── openai_compatible.go  # OpenAI, OpenRouter, Ollama, Groq, etc.
├── analyzer/                 # Docker image layer inspection
├── cli/                      # Cobra commands (analyze, init-config, mcp)
├── config/                   # YAML config loading with env var expansion
├── mcp/                      # MCP server (stdio transport, 4 tools)
├── samples/                  # Test Dockerfiles (light → nightmare)
└── web/                      # Landing page (React + Three.js)

📄 License

_{Built with 🐳 and a touch of sass.}

FilesExpand file tree

README.md

Latest commit

History