A second pair of eyes that reads first.
With your own Claude API key, or a model on your own machine, every container and the host terminal get a Claude tab. It reads logs, events, stats and inspect data and suggests a fix. What else the AI can do — your own model, acting within rules, other AI tools through MCP — is on the AI page.
Who is that in the corner?Read the comic: an advisor who reads first, and waits for your click →What you can do
On every container
And on the host terminal, where it reads what the terminal shows. It is told where the app and the engine run: WSL, Docker Desktop, a server or this computer.
Reads what matters
Logs, events, stats, inspect data and compose projects. That is all it sees.
Secrets masked
Passwords and tokens are masked before anything is sent.
You press the button
By default a proposed action, such as a restart, runs only when you click it.
Commands wait for Enter
A proposed command goes into the terminal prompt. You decide whether to run it; the AI never types there itself.
When a build fails
The build dialog, and a failed rebuild of a compose service, offer a diagnosis to copy, or a conversation that starts from the Dockerfile and the end of the build output.
Pick the model
Opus 5.5 by default for a stubborn problem; Sonnet 5.5, Fable 5.1 or Haiku 4.5 when quicker matters more. A conversation can switch to another model halfway.
Or your own model
Any address that speaks the OpenAI API: Ollama, Docker Model Runner, LM Studio. Then nothing has to leave your network.
Your key, stored safely
Encrypted on your computer and readable only by your user account.
Your own bill
Each question to Claude costs a little API usage, billed to your own Anthropic account. Settings shows the tokens per model of the last thirty days, also those of Ctrl+K. Give a model a price per million tokens to see an amount, and a monthly budget gives one notification when it is reached.
From the palette
Ask straight from Ctrl+K, without opening a tab first.
Hands, if you want
Let it act within rules you set, and let Claude Code in through MCP. Both are on the AI page.
Setup
Paste a key, or point it at your own model.
Open Settings and, under AI, paste an API key from the Claude Console and pick a model, or choose a model on your own machine. Leave both empty and the Claude tabs stay out of the way.
All features