← All Diagrams

Voice Drift: Problem → Solution

Before/After comparison showing the voice persistence problem and FAF solution

❌ Without FAF

Session 1 1,750 tokens

"Here's my project architecture..."

"Our code style is..."

"The team prefers..."

✓ AI learns your voice
↓
Session closes
↓
Session 2 1,750 tokens

"Here's my project architecture..."

"Our code style is..."

"The team prefers..."

⚠️ Everything forgotten
91%
Tokens wasted on re-discovery

✅ With FAF Voice API

faf init --xai
↓
One-time setup 150 tokens

📄 project.faf created

grok: voice: Leo

persona: eternal memory

retrieval_mode: hybrid

✓ Voice persisted forever
↓
Every session after
↓
All sessions 0 setup tokens

🚀 Grok reads project.faf

🧠 Voice remembered

⚡ Zero re-explain

✓ Eternal memory active
91%
Token reclaim achieved