You can try out the language by opening the interpreter and making queries to the language.

$ cosmos -i
> x=1
| x = 1
> x=1 or 2=x
| x = 1
| x = 2

Writing a file

Make a file with the content,

print('hello world')

The file can be loaded with the -l flag.

$ cosmos -l hello
'hello world'`


Instead of functions, Cosmos has relations.

They're made using the rel keyword.

//note that the there is no 'return' in the definition
//instead, the parameter y is explicit
//this is typically the 'output' parameter
rel double(x, y)
    y = x*2

double(4,x) //x is 8

Whereas functions have one output, relations may have zero, one or more outputs. You can check this by making queries at the interpreter.

cosmos -i
> x=1 or x=2 //this query has two answers (outputs)
| x = 1
| x = 2

If the system picks one answer and it turns out to be invalid, the system will backtrack and pick the other.

rel p(x)
    x=1 or x=2

rel main()
    io.writeln(x) //2

Relations may adopt function syntax. This lets relations be nested.

print(double(3)) //this will print 6

Logic-wise, double(4,x) is read as a statement: "the double of 4 is x".

double(4) reads as "the double of 4".

First-class relation

Relations are first-class values. It's possible to define a relation within another.

rel p(x)
    rel temp(x)
        x = 2

p(x) //x is 2


rel p(x)
    temp = rel(x)
        x = 2


Functors are composite data.

functor(F, Functor) //declares an object for creating functors
x = F(1, 2) //x is assigned to a functor F composed by the values 1 and 2
x = F(1, a) //uses pattern matching to match F(1, 2) against F(1, a)
print(a) //2

The special relation functor is used to declare F as an object for making functors.

Lists are syntax sugar for the functor Cons. Here are two ways to define a list:

l = [1, 2]
l = Cons(1, Cons(2, Cons))

Relations such as first, map and filter can be used to manipulate lists.

l = [1,2,3]
list.first(l, head) //head is 1, tail) //tail is [2, 3],, l2) //l2 is [2, 3, 4],, l) //l3 is [0, 1, 2]
list.filter(l, rel(x) x!=3;, l4) //l4 is [1, 2]


Variables are immutable. Instead of modifying a value we create a new one.

l2 = list.push(l, 55) //instead of modifying l, we create a new variable l2
io.writeln(l)  //[1, 2, 3]
io.writeln(l2) //[1, 2, 3, 55]

Cosmos adopts many principles and features that are common in functional programming languages (although the principles apply to relations rather than functions).


Cosmos manages a balance between strictness and non-strictness. Writing the type of a variable is (almost always) optional.

    Integer n = 7
    Real x = 5.2
    String s = 'abc'
    z = 5 //z is implied to be an Integer
    Functor l = [1, 2, 3]
    functor(F, Functor)
    Functor f = F('apple', 5)

The type system supports composite types.

    Functor String Number f2 = F('apple', 2)

Functor String Number is a composite type that accepts any functor whose first element is a string and second is a number.

    Relation Any Any p = double

Relations get composite types. Relation Any Any is a type that accepts any relation with exactly two arguments.


Tables (also known as maps, dictionaries, etc.) are structures that map keys to values.

Table t = {x=1 and y=2}
table.set(t, 'a', 1, t2)

print(t) //{'x': 1, 'y': 2}
print(t2) //{'x': 1, 'y': 2, 'a': 1}

Making a module

Cosmos files are typically given the .co extension.

rel p(x)

t = {
    'p' = p

As we have mentioned, Cosmos is in part inspired by imperative scripting languages--specifically, prototypal ones.

The principle is thus the same. The relation p is inserted into a table, which is then exported.

This could've been written as,

t = {
    rel p(x)

We then use the special relation require,

require('x', x)
print(x) //2


The language is whitespace sensitive.

rel p(x)

This could be a single line.

rel p(x) x!=1 and x<5;

It's possible to drop the whitespace semantics by writing the unnecessary characters, although this is not generally advisable.

Note that relations from different lines are separated by ands (semicolons are only used to end the indendation).


There is no boolean type. Instead, relations themselves are "booleans".

Using cases

As you may have noted, and/or is a huge part of the language. Hence, there are many operators that are shorthand for and/or. One of them is case (alias: cond).

    s = 'a'
    s = 'b'
    x = 1
    x = 2

This is sugar for,

(s = 'a') or (s = 'b' and x = 1) or x = 2


There's also support for if-statements.

if(s = 'a')
    x = 0
    x = 2

This is equivalent to,

(s = 'a' and x = 0) or (*s != 'a'* and x = 2)

Note that the condition is negated. This is as expected. The keyword 'else' implies the condition is not the case.

The downside of this is that more complex conditions may not be accepted by the compiler.


Negation is complicated

The condition this time is p(x). How do we negate an arbitrary relation like p(x)? All the while keeping the code logically pure?

As is known by now, the first logic programming language implemented negation in a very naive way. The result is that you could not be sure your code was sound and the operator is deprecated to this day.

Cosmos' philosophy is that if logically pure code is not supported, the compiler will give an error.

This may not seem like much, but it means you are free to use operators without worrying. When operators could affect the rest of the code in unseen ways before, this is an important change. It is much closer to the goal of "programming in logic".

As the main conditional operator of the language, if is guaranteed to be a logically sound conditional, or, if it can't do that, an error will occur.

if vs when


One way out of this is to replace if with the similar operator when. This is sugar for,

(p(x) and x = 1) or (x = 2)

Note that it does not negate the clause. As noted, this is a bit misleading since it still uses the 'else' keyword. Furthermore, the condition is kind of redundant. However, it evaluates to pure code.

In conclusion, when is an operator for when you do not need to negate the clause (or you want to negate it manually). It's simply a more procedural-looking way of writing 'case'.

Another reason for having when is that it can be easily switched with if if there is a need to.

Impure operators

While Cosmos puts emphasis on pure code, it may not be possible to avoid resorting to impure operators (specially in the language's current state).

These are not the worse LP has to offer (in fact, Prolog often calls them soft-cut, as they are mild versions of the cut operator), but the following operators are not completely pure:

//this is a conditional that will 'choose' one of the options without much concern for purity

//this will only select the first answer given by p(x)
once p(x)