I built a minimalist protocol to communicate with AI without ambiguity

The problem

Every developer who works with AI knows this frustration:

you write a long, detailed prompt, and the AI generates

something that's almost right but inconsistent,

over-engineered, or missing critical details.

The problem isn't the model. It's the language.

Natural language is ambiguous by design. What's obvious

to you has ten possible interpretations for an AI.

The solution: NEXUS

NEXUS is a tachygraphic protocol — structured shorthand

that compresses complex software architecture into

precise, unambiguous intent.

Instead of writing:

"Create a modern dashboard with authentication,

analytics charts, user management, real-time

notifications, dark mode, responsive design..."

You write:

@React @Tailwind

Page Dashboard

  @Auth[mode:jwt]

  Layout SplitView

  Section Analytics #glass

    Chart < MetricsData [paginate:10]

  Section Users

    Table < User !bold

      !error:404 -> /not-found

8 lines. Zero ambiguity. The AI knows exactly what

to build.

What NEXUS is (and isn't)

NEXUS is not a programming language. It doesn't

replace JavaScript, Python, or any other language.

It's the communication layer between human intent

and AI generation. You write NEXUS, the AI generates

the real code.

How it works

// Frontend

Page ProductDetail

  Layout Stack #gap-2

  Section Hero

    Image < product.thumbnail [ratio:16/9]

    Text < product.name !bold !xl

  Section Actions

    ( product.inStock ) ->

      Button "Add to Cart" => CartStore.add(product)

    :

      Badge "Out of Stock" #muted



// Backend

Model Order

  Entity id !pk

  Entity status default:pending

  Entity user -> Model.User

  Entity items -> Model.Product [many]



Controller OrderController

  Router ApiV1

    Endpoint POST /orders => OrderService.create()

      !error:400 -> /error/bad-request

      !error:500 -> /error/server

The library

nxlang is the open source TypeScript library that

implements the protocol:

• 
  validateNexus() — validates syntax in real time


• 
  buildSystemPrompt() — generates the AI context


• 
  createDefaultConfig() — project DNA configuration
  

npm install nxlang

135 tests passing. Library-first architecture so you

can embed it in your own tools and editors.

Beyond code

The three primitives of NEXUS are universal:

• 
  @ Directives — define context (@React, @Legal, @Scientific)


• 
  # Tokens — define labels (#urgent, #primary, #confidential)


• 
  -> => Relations — define flows and actions

The protocol was designed for code but the problem it

solves exists in every industry that uses AI.

Links

• 🌐 Website:
  


• 💻 GitHub: open-souse/Nexus
  

Would love to hear what the Dev.to community thinks —

especially if you've felt this same frustration with

long prompts to AI.