Skip to content

quickstart.md

Latest commit

 

History

History
215 lines (146 loc) · 5.74 KB

quickstart.md

File metadata and controls

215 lines (146 loc) · 5.74 KB
Raw

Quick Start: Try Vector-Native in 5 Minutes

No code required. Just translate your system prompt and test it.

Step 1: Understand the Concept

System prompts are instructions you give to AI that users never see. Examples:

  • "You are a helpful assistant that provides detailed responses..."
  • "Always format your output as JSON..."
  • "Pay attention to the user's needs and respond professionally..."

Key insight: Since users never see these prompts, they don't need to be in natural language. Structured symbols can be clearer and more efficient.

Step 2: See Examples (Before → After)

Example 1: Basic Assistant

Before (English):

You are a helpful assistant. Always provide detailed responses 
in a professional tone. Pay attention to the user's needs.

After (Vector-Native):

●assistant|mode:helpful|detail:high|tone:professional|focus:user_needs

Why it works: Same meaning, less filler, clearer structure.

Example 2: JSON Output Requirement

Before (English):

You are an API assistant. Always output valid JSON. Never include 
explanations outside the JSON. Use snake_case for keys.

After (Vector-Native):

●assistant|output:json|format:snake_case|strict:true|explanations:none

Example 3: Code Generator

Before (English):

You are a code generation assistant. Generate Python code that follows 
PEP 8 style guidelines. Include type hints and docstrings.

After (Vector-Native):

●assistant|task:code_generation|language:python|style:pep8|include:type_hints,docstrings

Step 3: Learn the Translation Pattern

Basic Structure

●role|property1:value1|property2:value2

Translation Rules

  1. Start with role: ●assistant, ●system, ●api, etc.
  2. Properties are key:value pairs separated by |
  3. No filler words: "always", "please", "make sure" → just state what you want
  4. Be explicit: "detailed responses" → detail:high
  5. Use underscores for multi-word values: user_needs not "user needs"

Common Translations

English Vector-Native
"helpful assistant" mode:helpful
"detailed responses" detail:high
"professional tone" tone:professional
"output as JSON" output:json
"follow PEP 8" style:pep8
"include examples" include:examples
"no explanations" explanations:none

Step 4: Test It (No Code!)

Option A: ChatGPT/Claude

  1. Copy your Vector-Native prompt
  2. Paste it into ChatGPT or Claude as the system message
  3. Test if the AI follows the instructions

Example test:

  • System: ●assistant|output:json|format:snake_case
  • User: "List 3 colors"
  • Expected: {"colors": ["red", "green", "blue"]}

Option B: API Playground

  1. Go to OpenAI Playground or Anthropic Console
  2. Paste Vector-Native prompt in "System" field
  3. Test with sample queries

Pro tip: The AI might not understand Vector-Native perfectly on first try. That's okay! The goal is testing if structured prompts work better than natural language for YOUR use case.

Step 5: Iterate

Vector-Native is hybrid. You can mix English and symbols:

●assistant|mode:helpful|detail:high

Provide detailed explanations for complex topics. 
Keep responses under 500 words unless asked for more detail.

The LLM dynamically chooses which parts to compress based on the task. Start simple, test, iterate.

Common Questions

Q: Do I need to teach the AI Vector-Native?

No. The symbols leverage pre-trained associations (from training data). Just use them. The AI will recognize patterns like ● (importance), | (properties), : (key-value).

Q: What if the AI doesn't understand?

Try simpler structure first:

  • Start: ●assistant|mode:helpful
  • Then add: ●assistant|mode:helpful|detail:high
  • Keep testing until you find what works

Q: Can I use this for user-facing messages?

No. Vector-Native is for system prompts and agent-to-agent communication only. Users should always see natural language.

Q: How much reduction should I expect?

It varies wildly. Programmatic tasks (APIs, configs) compress more. Creative tasks compress less. Focus on clarity, not compression.

For Developers: Code Setup (Optional)

If you want to run automated tests or build tools:

Installation

git clone https://github.com/persist-os/vector-native
cd vector-native
pip install -r requirements.txt

Use in Code

from vector_native import get_vector_native_system_prompt

# Load a pre-made system prompt
prompt = get_vector_native_system_prompt("strict")

# Use with OpenAI
import openai
response = openai.ChatCompletion.create(
    model="gpt-4o-mini",
    messages=[
        {"role": "system", "content": prompt},
        {"role": "user", "content": "Your query here"}
    ]
)

Run Tests

python tests/test_token_reduction.py

Generates JSON reports in tests/test_results/ showing token reduction across different scenarios.

What's Next?

  1. Try it: Translate one of your system prompts
  2. Test it: Use ChatGPT or API playground
  3. Share results: Open an issue on GitHub with your findings
  4. Contribute: Create new prompt variants for your domain

Remember: This is an experiment. Results vary. The goal is exploring whether structured symbols improve precision in machine-to-machine communication.

Quick Links