25 built-ins, all on device

current_datetime

Returns the current date, time and time zone on this device. Called only when the answer cannot be given without today's date.

No parameters

calculate

Works out an arithmetic expression exactly, for example 17*23 or (4.5+2)/3. Parsed by a hand-written recursive-descent evaluator rather than NSExpression, which raises exceptions Swift cannot catch.

expression:string

Arithmetic using + - * / % ^ ( ) and decimal numbers

convert_units

Converts a number from one unit to another, for example 5 km to miles. Never invents a conversion the user did not ask for.

value:number

The number to convert

from:string

Source unit such as km, mi, kg, lb, c, f, l, gal, mph, gb

to:string

Target unit

search_conversations

Searches your earlier Sotto conversations for a phrase and returns matching snippets. Everything stays on the device — but it reads your chats, so it asks first.

query:string

Words to look for

text_statistics

Counts the characters, words and sentences in a passage you supplied. Ships off: the system model called it on prompts like “say hello in one word”, where a word count is noise rather than help.

text:string

The text to measure

date_difference

Measures the gap between two dates in days, weeks, months and years.

from:string

Start date as YYYY-MM-DD, or today, tomorrow, yesterday

to:string?

End date; the current moment when left out

date_shift

Gives the date a number of days, weeks, months, years, hours or minutes before or after another date.

amount:number

How far to move; negative goes backwards

unit:string

days, weeks, months, years, hours or minutes

date:string?

Starting date; today when left out

time_in_zone

Reports the current date and time in another time zone and how far ahead or behind this device it is.

zone:string

A city or identifier such as Tokyo or Asia/Tokyo

calendar_facts

Reports the weekday, month length, quarter, day of the year, ISO week number and leap year for one date.

date:string?

The date to describe; today when left out

transform_text

Rewrites text in a chosen style: upper, lower, title, sentence, snake, kebab, camel, pascal, slug, trim, reverse or strip_accents.

text:string

The text to rewrite

style:string

One of the styles listed in the description

replace_text

Replaces every occurrence of one exact phrase with another and says how many changed. Matches literally — a model cannot supply a regular expression that backtracks catastrophically.

text:string

The text to change

find:string

The exact phrase to look for

replace:string?

What to put in its place; empty deletes it

match_case:boolean?

Match capitalisation exactly; true when left out

extract_matches

Pulls every email, url, number, hashtag, mention, ip_address, date or phone number out of text. Eight fixed, audited patterns — never one the model wrote.

text:string

The text to search

kind:string

email, url, number, hashtag, mention, ip_address, date or phone

limit:number?

How many to list, 1 to 100; 25 when left out

sort_lines

Sorts the lines of a list ascending, descending, longest, shortest or reversed, and can drop repeats.

text:string

The lines to sort, one per line

order:string?

ascending, descending, longest, shortest or reverse

unique:boolean?

Remove repeated lines

word_frequency

Counts which words appear most often in a piece of text.

text:string

The text to count

limit:number?

How many words to list, 1 to 50; 10 when left out

ignore_common:boolean?

Skip words like the and of; true when left out

compare_texts

Compares two texts line by line and lists the lines that differ.

first:string

The original text

second:string

The text to compare against it

format_json

Checks that JSON is valid and rewrites it pretty-printed or minified, naming the fault when it is not.

json:string

The JSON text

style:string?

pretty or minified; pretty when left out

query_json

Reads one value out of a JSON document by dot path, such as user.address.city or items.0.name.

json:string

The JSON text

path:string

Dot path to the value; empty returns the whole document

summarize_csv

Reports the rows, columns and per-column figures of CSV text.

csv:string

The CSV text

delimiter:string?

Single character between fields; a comma when left out

has_header:boolean?

First row holds column names; true when left out

describe_numbers

Gives the count, sum, mean, median, minimum, maximum, range and standard deviation of a list of numbers.

numbers:string

The numbers, separated by commas or spaces

percentage

Works out percentages: percent_of, what_percent, change between two values, or increase and decrease by a percent.

mode:string

percent_of, what_percent, change, increase or decrease

value:number

The first number

other:number

The second number, or the percentage for increase and decrease

convert_base

Converts a whole number between bases 2 to 36, such as decimal to binary or hexadecimal.

value:string

The number, written in from_base

from_base:number?

Base it is written in, 2 to 36; 10 when left out

to_base:number?

Base to convert to, 2 to 36; 2 when left out

encode_text

Encodes or decodes text as base64, hex or url percent-encoding. These encodings are reversible by anyone and are not security.

text:string

The text to convert

format:string

base64, hex or url

direction:string?

encode or decode; encode when left out

hash_text

Computes the SHA-256, SHA-384 or SHA-512 digest of text. Not for encrypting anything or storing a password.

text:string

The text to hash

algorithm:string?

sha256, sha384 or sha512; sha256 when left out

random_number

Returns up to twenty random whole numbers in a range, for dice, picks and draws. Never called to invent a figure for an answer.

minimum:number?

Lowest possible number; 1 when left out

maximum:number?

Highest possible number; 100 when left out

count:number?

How many numbers, 1 to 20; 1 when left out

estimate_tokens

Estimates how many tokens a piece of text costs a model, alongside its characters and words.

text:string

The text to measure