Commands & doctor
Commands
| Command | What it does |
|---|---|
kaja config paths | print where every config file resolves on this machine. Start here when unsure which file Kaja reads |
kaja config wizard | re-run the setup wizard |
kaja config diff | show what fetch would change, without writing anything |
kaja config fetch | rewrite models.toml and commands.toml from the defaults (backing them up if they differ), and write secrets.toml if you have none |
kaja doctor | test every key, model and tool, see below |
kaja config fetch takes models.toml and commands.toml from the Kaja server’s defaults, or from the templates bundled in the binary when you’re offline or pass --offline. Your own patterns in commands.toml’s custom survive a fetch. secrets.toml comes from the bundled template with every section commented out, and is only written if you have none, so fetching never loses a key. --only models, --only commands or --only secrets limits it to one file. It never touches settings.toml or the marketplace/ folder. Use it to pick up new defaults after an upgrade, or to recover a broken file.
Checking keys and models
kaja doctor tests every credential your config relies on: model providers, HTTP tools and MCP servers, the Telegram token and web search. In a terminal it asks for anything missing or failing, tests the new value, and saves it to secrets.toml. A value that fails is saved too, so you can fix it there, and stays on the to-do list until it works.
It then tries every model by task, and shows each chat model’s context window and where the number came from (models.toml, detected from the server, or assumed). When a task’s model stops answering and a model further down that lists the task works, it offers to switch: the broken models ahead of it are commented out of models.toml, so the working one comes first. It ends with what’s still broken, and lists every tool by origin plus anything left out and why.
Next: