The .nt syntax

The .nt format is meant to be easy to read and write. This page is a quick reference to all its rules.

The rules

• Indent with spaces. Use two spaces per level. Do not use tabs.
• Comments start with #. Anything after # on a line is ignored.
• Settings are name: value. One per line.
• Lists use -. One item per line, or write them inline like [a, b, c].
• Long text uses |. Everything indented under it is kept as-is.
• Secrets use env(NAME). This reads a value from your environment.

# this is a comment
agent age
  description: Guesses a person's age.       # a note
  model: anthropic/claude-sonnet-5
  tools: [current_year, fs_write]            # inline list
  skills:                                     # list on separate lines
    - estimation
  instructions: |                             # long text
    Guess the person's most likely age
    from the clues you are given.

The building blocks

Every .nt file is made of blocks. Each starts with a keyword and, for most, a name. Each has its own guide:

Reading secrets with env()

env(NAME) reads a value from your environment. You can give a fallback for when it is not set with env(NAME, fallback):

api_key: env(ANTHROPIC_API_KEY)
api_key: env(OLLAMA_API_KEY, ollama) # uses "ollama" if the variable is missing

Input and output types

input and output are lists of fields. Each field has a type. You can write it short:

input:
  clues: string

Or with a description:

output:
  age:
    type: number
    description: The best single guess.
  reason:
    type: string
    description: One short sentence.

The types are string (text), number, boolean (true or false), and object.