Skip to content

test_context_aware.md

Latest commit

 

History

History
124 lines (104 loc) · 3.47 KB

test_context_aware.md

File metadata and controls

124 lines (104 loc) · 3.47 KB
Raw

Context-Aware .m8 Loading Examples

Scenario 1: General Browsing (No Context)

You: "What projects do we have?" Smart Tree loads: Minimal - just essences

• Smart Tree - AI-optimized directory visualization
• 8b.is website - Company portal
• MEM8 - Wave-based memory system

Tokens used: ~50

Scenario 2: Talking About Websites

You: "How's the 8b.is website coming along?" Smart Tree detects: "website" keyword Auto-expands: 8b.is/.m8 to medium detail

📂 8b.is (88.8Hz)
  8b.is website - Company portal for 8-bit inspired AI services
  Keywords: 8b.is, website, portal
  📁 Children:
    • frontend (92.3Hz)
    • api (87.5Hz)
    • docs (45.2Hz)

Tokens used: ~150

Scenario 3: Specific Technical Discussion

You: "The 8b.is website API needs authentication" Smart Tree detects: "8b.is", "website", "API" - HIGH RELEVANCE! Auto-expands: Full detail + drills into api/

╭──────────────────────────────────────────────────
│ 📂 8b.is
│ 🌊 Frequency: 88.8Hz
│ 📝 8b.is website - Company portal for 8-bit inspired AI services
│ 🏷️ Keywords: 8b.is, website, portal
│ 📁 Children:
│   • frontend (92.3Hz)
│   • api (87.5Hz) ← AUTO-EXPANDING THIS
│   • docs (45.2Hz)
│ 🎯 Context Triggers:
│   website → frontend/
│   API → api/ ← TRIGGERED!
╰──────────────────────────────────────────────────

📂 8b.is/api/.m8 (87.5Hz)
  RESTful API with wave-based authentication
  Endpoints: /auth, /memories, /waves
  Port: 28428
  Auth: JWT with MEM8 signature

Tokens used: ~400 (but ONLY when needed!)

Scenario 4: Smart Tree Development

You: "The tokenizer in Smart Tree needs work" Smart Tree detects: "tokenizer" trigger Auto-loads: src/tokenizer.rs context

📍 Drilling down: src/tokenizer.rs
  Tokenization system - 90% compression
  Patterns: node_modules→0x80, .rs→0x91
  Methods: tokenize(), decode(), compression_ratio()

The Magic: Progressive Loading

Without Context-Aware Loading:

  • Load everything: 10,000+ tokens
  • Most irrelevant to current discussion
  • Context window fills quickly

With Context-Aware .m8:

  • Base load: 50 tokens (just essences)
  • Relevant expansion: +100-400 tokens
  • Total: 150-450 tokens (95% reduction!)

Real Example:

Starting context (50 tokens):

Projects: smart-tree, 8b.is, mem8, marqant

You mention "memory":

+100 tokens: MEM8 expanded with wave frequencies

You mention "binary format":

+150 tokens: .m8 format specs loaded

You mention "tokenization":

+200 tokens: Smart Tree tokenizer details

Total: 500 tokens vs 10,000 tokens without context awareness!

Implementation in Smart Tree:

// In scanner.rs when encountering .m8 files
if path.ends_with(".m8") {
    let keywords = extract_conversation_context();
    let content = context_reader.load_contextual(&path, &keywords)?;

    // Only expand if relevance > threshold
    if relevance > 0.7 {
        // Drill down automatically
        expand_children(&path, &keywords)?;
    }
}

Frequency-Based Relevance:

  • High frequency (>150Hz): Hot zones, recent work
  • Medium (50-150Hz): Active projects
  • Low (<50Hz): Documentation, archives

The .m8 frequency helps determine expansion priority!