JibL

A strongly-typed, strictly procedural, AI-native programming language with trilingual keywords — English, Français, and Español.

JibL source compiles to a clean S-expression IR that a tiny C runtime evaluates. The IR is human-readable and designed to be inspected, generated, and transformed by AI tools.

code english

func int fibonacci(int n) returns int [:
    if n <= 1 [:
        return n
    :]
    return fibonacci(n - 1) + fibonacci(n - 2)
:]

int i = 0
while i < 10 [:
    print("fib(" + i + ") = " + fibonacci(i))
    set i = i + 1
:]

---

Features

• Trilingual keywords — write code in English, Français, or Español
• Strongly typed — int, float, string, bool, result<T,E>, T[], json, ai_response
• Block scope — variables are scoped to [: :] blocks; shadowing is a compile-time error
• No pointers — safe by design; no manual memory management exposed to the user
• Explicit error handling — result<T,E> type with ok(v) / error(msg) constructors
• Contracts — requires / ensures checked at runtime on every function call
• AI built-ins — ask("prompt") queries an LLM at runtime; @ai "..." generates function bodies automatically and caches them
• S-expr IR — --emit dumps the compiled program as S-expressions for AI inspection and tooling
• Standard library — file I/O, HTTP (GET/POST), JSON parsing

---

Requirements

Dependency	Notes
C11 compiler (cc / gcc / clang)	Standard on macOS and Linux
libcurl	For HTTP and AI API calls. macOS: brew install curl. Linux: apt install libcurl4-openssl-dev

---

Build

git clone https://github.com/jib63/JibL.git
cd JibL
make

The jibl binary is created in the project root.

make run          # runs examples/hello_en.jibl
make emit         # prints S-expr IR for hello_en.jibl
make clean        # removes build artifacts
make clean-cache  # clears the @ai function cache

---

Usage

./jibl <file.jibl>          # compile and run
./jibl --emit <file.jibl>   # print S-expr IR, do not execute

---

Language Reference

File header (mandatory)

Every .jibl file must begin with a language declaration:

code english
code français
code español

Types

Type	Description
int	64-bit signed integer
float	64-bit double
string	UTF-8 string
bool	true / false
T[]	Homogeneous array of type T
result<T, E>	Either ok(value) or error(reason)
json	Parsed JSON object
ai_response	Response from an LLM (.content, .model, .tokens)
void	No value (function return only)

Variables and constants

int x = 10                      @@ mutable
const string TAG = "v1"         @@ read-only — set is a compile error

Assignment

set x = x + 1

Control flow

if x > 5 [:
    print("big")
:] else [:
    print("small")
:]

while x > 0 [:
    set x = x - 1
:]

Functions

func int add(int a, int b) returns int [:
    return a + b
:]

int total = add(3, 4)       @@ expression call
call greet("World")         @@ statement call (return value discarded)

Error handling with result

func result<int, string> safe_div(int a, int b) returns result<int, string> [:
    if b == 0 [:
        return error("division by zero")
    :]
    return ok(a / b)
:]

result<int, string> r = safe_div(10, 0)
if r.ok [:
    print("answer: " + r.value)
:] else [:
    print("error: " + r.error)
:]

Contracts

func int divide(int a, int b) returns int
    requires b != 0
[:
    return a / b
:]

requires is checked before the function body executes. ensures is checked before return, with __result bound to the return value.

Arrays

int[] scores = [10, 20, 30]
set scores = append(scores, 40)
int first = scores[0]
int size = len(scores)

Comments

@@ single-line comment

@@@
    multi-line
    comment
@@@

Multiline strings

string msg = """
    Hello,
    World!
"""

---

AI Features

ask() — runtime LLM query

ai_response r = ask("What is the boiling point of water in Celsius?")
print(r.content)
print(r.model)
print(r.tokens)

@ai — AI-generated function body

@ai "implement bubble sort for a int array, return sorted copy"
func int[] bubble_sort(int[] arr) returns int[] [:
:]

On first run the AI generates a (block ...) S-expr body and caches it in .jibl_cache/. Subsequent runs load from cache — no network needed.

Configuration

export JIBL_AI_PROVIDER=anthropic   # or: openai
export JIBL_AI_KEY=sk-ant-...
export JIBL_AI_MODEL=claude-sonnet-4-6   # optional, has defaults

---

Keyword Reference

Concept	English	Français	Español
if	if	si	si
else	else	sinon	sino
while	while	tantque	mientras
function	func	fonc	func
return	return	retour	retorno
returns (sig)	returns	retourne	retorna
assignment	set	set	set
print	print	afficher	mostrar
call (stmt)	call	appel	llamar
requires	requires	necessite	requiere
ensures	ensures	garantit	garantiza
ask (AI)	ask	demander	preguntar
append	append	ajouter	agregar
length	len	longueur	longitud
true	true	vrai	verdadero
false	false	faux	falso
and	and	et	y
or	or	ou	o
not	not	non	no
int type	int	entier	entero
float type	float	decimal	decimal
string type	string	chaine	cadena
bool type	bool	booleen	booleano
void type	void	vide	vacio
result type	result	resultat	resultado
ai_response type	ai_response	reponse_ia	respuesta_ia

code, const, set, [:, :], @@, @@@, """, @ai, ok, error, and all stdlib function names are the same in every language.

---

Examples

File	What it demonstrates
examples/hello_en.jibl	Hello World in English
examples/hello_fr.jibl	Hello World in Français
examples/hello_es.jibl	Hello World in Español
examples/fibonacci_en.jibl	Recursion and while loop
examples/contracts_en.jibl	requires contract
examples/result_type_en.jibl	result<T,E> error handling
examples/arrays_en.jibl	Arrays, append, len, indexing
examples/ai_ask_en.jibl	ask() runtime AI query

---

License

MIT — see LICENSE.