Skip to main content

​
What is the Digital Twin?

FERAL’s Digital Twin is a cognitive replica that can answer questions “as the user” by drawing on their full memory corpus, identity files, knowledge graph, and personality profile.

​
How It Works

  1. Identity — Loaded from ~/.feral/USER.md (name, personality traits, priorities).
  2. Memory — Recent episodes and semantic search over long-term memory.
  3. Knowledge Graph — Structured triples (subject → predicate → object) extracted from conversations.
  4. LLM — Synthesizes all context to answer “as you would.”

​
Ethical Bounds — Act-as-User

The Digital Twin operates under strict ethical boundaries:
  • Read-only by default — The twin can answer questions and predict preferences but cannot take actions on behalf of the user without explicit approval.
  • No impersonation — The twin must never be used to communicate with others pretending to be the user.
  • Transparency — Any output from the twin should be clearly labeled as AI-generated.
  • User control — The user can disable the twin at any time via settings.json → features.digital_twin_enabled: false.
  • Data locality — All memory and identity data stays on the user’s device. The twin never sends personal data to external services beyond the configured LLM API.

​
Troubleshooting

  • Twin returns generic answers — Ensure ~/.feral/USER.md exists with personality info. More memory = better answers.
  • “wasn’t able to reason” error — LLM API is down or misconfigured. Check OPENAI_API_KEY or your configured provider.
  • Slow responses — The twin queries memory + KG + LLM sequentially. Check network latency to your LLM provider.
  • Predictions low confidence — Not enough relevant memories. Use FERAL more to build up context.
  • Checking logs — feral.digital_twin logger. Set FERAL_LOG_LEVEL=DEBUG for prompt details.