AI-NATIVE · OPEN SOURCE · BUILT IN NIGERIA

The language where
think is a keyword

NEKOVA is a programming language with AI reasoning built into its grammar — not bolted on as a library. Simple enough to teach the classmate who gave up on Python. Serious enough to ship real programs.

Get Started Read the Tour

SCROLL

Chapter One — The Question

It started with a classmate who quit.

NEKOVA's creator, Emmanuel King Christopher, was teaching himself to code in Nigeria when he watched a classmate give up on programming entirely — not because the ideas were too hard, but because the error messages were. A missing colon. An off-by-one indent. Python's own honest, technically-correct exceptions, dropped on a beginner with zero context.

The problem wasn't Python. Python is a genuinely well-designed language. The problem was that nobody had built a language whose first priority was the moment things go wrong — the moment a learner is most likely to close the laptop for good.

error message, before and after

$ python broken.py

IndentationError: unexpected indent (broken.py, line 4)

$ nekova run broken.nk

error[E010]: Inconsistent Indentation

This line is indented 6 spaces, but that doesn't match

any enclosing block. Valid indent levels here: 0, 4

Tip: mixing tabs and spaces is a common cause.

That single idea — an error message's job is to teach, not just to be technically correct — became NEKOVA's founding principle. Everything else followed from it.

Chapter Two — The Idea

What if the language could think too?

By the time NEKOVA's design solidified, large language models had become something any program could reasonably lean on. So the second question followed naturally: if a language is being built from scratch anyway, why should "call an AI model" require an SDK, an API key dance, and a dozen lines of boilerplate?

NEKOVA's answer was to make it a keyword. Not a library. Not an import. A first-class part of the grammar, sitting right next to if and return.

task summarize(document):
    let result = think "Summarize this in one sentence: {document}"
    return result

No client to instantiate. No response object to unwrap. think asks, and hands back an answer — text, a list, a boolean, structured JSON, whatever the task needs — with the same directness as any other expression in the language.

What NEKOVA is built on

Three ideas, held to consistently

◆

AI-native, not AI-bolted-on

think, remember, recall, prompt, and observe are real syntax — checked by the parser, understood by tooling, testable with mock think. Not a wrapper around someone else's SDK.

▚

Errors that teach

Every error message states what happened, why, and what to check next — never a bare stack trace. Near-miss variable suggestions, exact recursion depth, honest uncertainty when the interpreter genuinely doesn't know the cause.

▲

Built to outgrow its own scaffolding

NEKOVA runs on Python today. The roadmap's endpoint is self-hosting — a lexer, parser, and interpreter written in NEKOVA itself. Every phase is scoped with that horizon in view.

See it end to end

A small program, start to finish

Destructuring, keyword arguments, a typed task, and an AI call — the everyday shape of a NEKOVA program.

enum Priority: LOW, MEDIUM, HIGH

task triage(title, priority=Priority.MEDIUM):
    let urgency = think f"Rate urgency 1-10 for: {title}" as number when error: 5
    return {"title": title, "priority": priority, "urgency": urgency}

let {title, urgency} = triage("Server down", priority=Priority.HIGH)
show f"{title} — urgency {urgency}/10"

Try it in the next sixty seconds

One install command. No API key required to start — NEKOVA ships with a mock AI provider for learning and testing.

$ pip install nekova-lang

Start the Quickstart View Source

The language wherethink is a keyword