mise

A CLI that provides a suite of AI-assisted tools to help you manage your Tandoor recipes. LLMs are not perfect, and mise will certainly get some things wrong, but if you have hundreds or thousands of recipes, it can help you manage them semi-autonomously.

Features

• Normalize messy ingredient rows so amount, unit, and food land in the right fields, reusing Tandoor's existing foods and units.
• Assign keywords to a recipe following your own custom keyword schema.
• Manage your local AI models, tracking downloads, disk usage, and unused weights for Ollama.

Install

go install github.com/derethil/mise@latest

Or build from source:

git clone https://github.com/derethil/mise.git
cd mise
go build -o mise .

A Nix flake is also provided (nix build .#mise).

Configuration

mise reads a config file from $XDG_CONFIG_HOME/mise/config.toml - see the example config for an example. You can also configure mise using env vars e.g. MISE_TANDOOR_TOKEN or with a global flag e.g. --tandoor-url.

Environment variables use MISE_, then the section and key in uppercase; for example, tandoor.token becomes MISE_TANDOOR_TOKEN.

Providers

Ollama is the only supported provider right now. small/large let you pick different models for lighter vs. more demanding tasks or you can provide a --model flag to override the model for one run.

Commands

• mise configure - manage configuration file
• mise recipe normalize - cleans up a recipe's ingredients by ensuring each amount/unit/food lands in the correct Tandoor field
• mise recipe keyword - assigns keywords to a recipe according to the keyword schema in your schema file
• mise recipe backup / restore - snapshot a recipe's JSON before you mess with it, or roll back to a snapshot later.
• mise models - manage the Ollama models mise uses
• mise logs - print the path to the current mise log file

Global flags:

• -m, --model - override the AI model to use for a single command
• -v, --verbose - enable debug logging (-vv for verbose output from the underlying AI library)
• -c, --config - override config file path

Keyword schemas

mise recipe keyword has no built-in keyword assignment system, so you can tell mise to follow whatever schema you prefer. It reads your schema file ($XDG_CONFIG_HOME/mise/keyword_schema.md by default), injects it into the prompt, and uses it to decide which keywords to add to a recipe.

## Cuisine

Assign the broad cuisine (Chinese, Japanese, American, French) and, when the
dish belongs to a distinct regional cooking tradition, that region as well
(Szechuan, Cajun, Sicilian). A region never appears without its broad cuisine.

## Dish type

Assign exactly one: Main, Side, Bread, Appetizer, Breakfast, Dessert, or Drink.

Each ## heading is one "keyword category". You can tell the model to return multiple keywords for a category if needed but mise makes a separate model call per category to improve accuracy. Anything above the first heading is included in every call. A schema with no headings is simply one call.

Existing Tandoor keywords are also shown to the model so it typically reuses them instead of creating a similar keyword.

A more fully-featured keyword classification system can be seen in this example schema file.

License

MIT, see LICENSE.