runcor-R++

Stop guessing.
Start specifying.

R++ is a structured specification language for LLMs. Write a script with typed blocks, named values, and explicit assertions. The LLM produces exactly what it describes.

Domain-agnostic. Works with Claude, GPT, Gemini, or any LLM.

▶ Try the Playground View on GitHub

The Problem

Natural language prompts are ambiguous

An LLM reading prose has to guess at intent, fill in gaps, and make assumptions. Each guess is a chance to get it wrong. R++ eliminates those gaps.

✕ Prose prompt

Build me a REST API for order management.
It should have CRUD endpoints. Only admins
can delete. Add pagination with a default
page size of 20 and max of 100. Use proper
error handling and auth on all routes.

The LLM has to guess:
  → What framework?
  → What auth mechanism?
  → What error format?
  → What status codes?
  → Role names? Route paths?
  → Dozens of implicit decisions…

✓ R++ script

TARGET {
  output: "Order management REST API"
  lang: Python + FastAPI
  profile: api
}
TOKENS {
  pageSize:20 | maxPageSize:100
  roleAdmin:"admin" | roleUser:"user"
}
BEHAVIOR {
  MUST validate input BEFORE handler
  MUST NOT allow DELETE UNLESS role === roleAdmin
}
CHECKLIST {
  [ ] MUST implement EXACTLY 5 endpoints
  [ ] MUST use TOKENS (NEVER raw literals)
}

Every value named. Every assertion explicit. Zero guesswork.

Language Overview

10 blocks. Any domain.

R++ is domain-agnostic. The same block structure can specify a React component, a REST API, a CLI tool, a data pipeline, or any artifact an LLM can generate.

TARGET

Declare what is being built and which domain profile applies.

TOKENS

Define named constants. Any repeated value gets a name.

FORMAT

Define formatting and transformation functions.

MAP

Define lookup tables. Map keys to values.

DATA

Provide input data. Use EXACT for verbatim output.

INIT

Declare state, computed values, filters, sorts.

STRUCTURE

Describe how the output is composed as a tree.

COMPONENT

Describe what each unit of output contains.

BEHAVIOR

Describe logic, events, flows, constraints.

CHECKLIST

Assert what must be true. Boolean verification.

Syntax

Structured, not verbose

Each block has a clear role. Profiles add domain shorthands without polluting the core.

Interactive

Try R++ live

Describe what you want to build and hit Run. Your prompt gets converted into a structured R++ spec you can use with an AI to run on runcor or develop an app of your own.

Input Text

R++ Output

Click ▶ Run to generate output…

Open full screen playground at rpp.runcor.ai

Domain Profiles

Core is universal. Profiles add vocabulary.

Activate a profile in the TARGET block. The core language is always available.

Precision

Every keyword is binary

No "should", "try to", or "ideally". Every enforcement keyword has a yes/no answer.

MUST

Required. Non-negotiable.

MUST NOT

Prohibited. Hard block.

ALWAYS

Every instance, no exceptions.

NEVER

Zero instances.

ONLY

Exactly this, nothing else.

EXACTLY

Precise count or value.

AT LEAST

Lower bound.

AT MOST

Upper bound.

UNLESS

Exception to a rule.

BEFORE

Ordering: this first.

AFTER

Ordering: this second.

PREFER

Soft preference.

FALLBACK

Recovery when primary fails.

OVERRIDE

Replaces a prior rule.

Stop guessing.Start specifying.