Token Optimization

How ÆtherLight helps you use AI more efficiently through patterns, conceptual breadcrumbs, and agent routing.

Why Tokens Matter

AI models charge by tokens (roughly 1 token ≈ 0.75 words). Every character in your prompt and response counts. The more efficiently you can communicate context to AI, the less you spend.

The Patterns Approach

Instead of dumping entire files into context, ÆtherLight uses patterns - structured knowledge that contains only what's relevant to that specific pattern.

What is a Pattern?

A pattern is a self-contained piece of knowledge about how something works in your codebase. It includes only the information needed to understand and work with that pattern - nothing more.

Conceptual Breadcrumbs

Patterns work through conceptual breadcrumbs - references that trigger AI to access pattern knowledge only when needed:

  • →Pattern names: Reference a pattern by name, AI knows what it means
  • →Chain of thought triggers: Comments that prompt AI to recall relevant context
  • →On-demand loading: Pattern details are only loaded when actually needed

This is similar to skills but more like a knowledge base - information that's available but not constantly in the context window.

Different from RAG

This approach is different from traditional RAG (Retrieval-Augmented Generation):

Traditional RAG

  • • Retrieves chunks of text based on similarity
  • • Often returns irrelevant context
  • • Adds everything to the prompt

Conceptual Breadcrumbs

  • • References patterns by concept/name
  • • AI decides when to access details
  • • Stays out of context until needed

Agent Routing Saves Tokens

Another way ÆtherLight saves tokens is through specialized agent terminals:

No More Context Switching

  • • Name your terminals as specialized agents (UI Agent, Backend Agent, API Agent, etc.)
  • • Route tasks from your sprint to the appropriate agent
  • • Each agent stays focused on its domain
  • • No token waste from constantly switching contexts in one terminal

See Terminal Context for details on setting up agent routing.

Tips for Efficiency

  • 1.Use sprint tasks - Pre-defined tasks have focused context
  • 2.Reference patterns by name - Let AI load details when needed
  • 3.Use agent routing - Route tasks to specialized terminals
  • 4.Be specific - "Fix login validation" beats "fix the app"