Skip to content

index.md

Latest commit

 

History

History
276 lines (200 loc) · 6.36 KB

index.md

File metadata and controls

276 lines (200 loc) · 6.36 KB

SNS - Shorthand Notation Script

Intuitive token-efficient notation for AI-to-AI communication

What is SNS?

SNS is not a programming language. It's a shorthand notation system that LLMs naturally understand without any training or documentation.

Think of it like:

  • 📝 Taking notes in shorthand
  • 🎵 Musical notation (musicians read it intuitively)
  • 🧮 Mathematical notation (everyone knows x → y)

Key insight: LLMs already understand these concepts. We're just making them shorter.

Quick Example

Before (Natural Language)

You are an AI assistant that analyzes user queries. Please carefully examine 
the following user question and extract the main keywords. Then, classify the 
user's intent into one of these categories: informational, complaint, or 
procedure. After that, expand the query into relevant search terms that could 
help find information. Finally, return the results in a structured format.

Token count: ~150 tokens

After (SNS)

q → kw_extract → kw
q → classify(intent_cats) → intent
kw + q → expand_q → terms
→ {kw, intent, terms}

Token count: ~30 tokens
Savings: 80% reduction 🎉

Why SNS Works

LLMs already understand:

  • → means "transform to" or "flows to"
  • kw is obviously "keywords"
  • classify() is self-explanatory
  • + means combine/merge

No training needed. No documentation needed. Just write it and it works.

Core Principles

  1. Intuitive shortcuts - If it feels right, it probably works
  2. Token efficiency - Every character saved = cost reduced
  3. Flexible creativity - Use emojis, symbols, abbreviations
  4. Self-documenting - Readable by humans and LLMs
  5. No formal grammar - It's notation, not code

Quick Start

Pattern 1: Flow Operations

input → operation → output

Examples:

query → analyze → insights
text → normalize → clean_text
docs → rank → top_results

Pattern 2: Pipelines

data | step1 | step2 | step3

Examples:

candidates | score | filter | sort | top(5)
text | lower | trim | tokenize

Pattern 3: Conditionals

condition ? true_action : false_action

Examples:

score > 0.7 ? keep : discard
results.empty ? expand_search : return_results

Pattern 4: Modifiers

+boost, -penalty, *emphasize, ~fuzzy

Examples:

results +boost(recency) +boost(local)
query ~match(docs)
score *2

Real-World Usage

RAG Orchestrator (Full Example)

Traditional Prompt (200 tokens):

You are the orchestrator in a RAG system. Analyze the user query to extract 
keywords, determine the intent, expand the query into search terms, and infer 
relevant categories. Return a structured object with these fields...

SNS Version (45 tokens):

# RAG Orchestrator
q → kw_extract → kw
q → classify(["info","complaint","procedure"]) → intent
(kw + q) → expand_q → search_terms
intent → infer_cats → categories

→ {
  search_terms,
  categories, 
  intent,
  kw
}

Result: Same output, 77% fewer tokens

Token Savings Across Operations

Operation Natural Language SNS Savings
Keyword extraction 45 tokens 12 tokens 73%
Classification 38 tokens 15 tokens 61%
Query expansion 52 tokens 18 tokens 65%
Ranking & filtering 67 tokens 22 tokens 67%
Full RAG pipeline 200 tokens 45 tokens 77%

Average savings: 68%

Documentation

Core Concepts

Operations

Examples

Experimental

Is SNS Right for You?

✅ Perfect For:

  • AI-to-AI communication (RAG stages, agents)
  • Internal LLM prompts (not user-facing)
  • Token-sensitive applications (cost optimization)
  • Multi-stage pipelines (orchestration)
  • Repeated operations (templates)

❌ Not For:

  • User-facing content (use natural language)
  • One-off simple prompts (overhead not worth it)
  • Creative writing tasks (flexibility needed)
  • When tone/empathy matters

The Magic: It Just Works

You don't need to teach LLMs how to parse SNS. They already understand:

# You write this:
q → analyze → {kw, intent, score}

# LLM understands:
"Analyze the query and return an object with keywords, intent, and score"

No special instructions needed. No system prompts explaining SNS. Just works.

Get Started

  1. Read the Philosophy to understand the approach
  2. Learn Core Patterns - 5 minute read
  3. Explore Examples to see it in action
  4. Start using SNS in your prompts immediately

Community

SNS is an open notation system. Contributions welcome:

  • New patterns that work
  • Creative shortcuts
  • Real-world examples
  • Token savings reports

Status: v1.0 - Active development
License: Open notation system
Created: October 2025

Quick Reference

# Flow
input → operation → output

# Pipeline  
data | step1 | step2 | step3

# Conditional
condition ? yes : no

# Loop (implied)
while x < 5: do_thing()

# Assignment (implied)
result = operation(input)

# Composition
(a + b) → process → output

# Modifiers
+boost -penalty *emphasize ~fuzzy

# Collections
[list] {object} (tuple)

# Functions
fn_name(args) → result

Ready to save tokens? Start with Core Patterns!