Growing Familiar's Mind
25:00
Brain XP: 0
Stay focused. Your familiar is learning while you work.
Beat 0 points!
Familiar Lost
Cause: energy depletion
Score
0
Level
1
Age
0s
Personality
Balanced
Mood
neutral
Use this as a quick map of how behavior, chat, and voice settings affect your familiar.
Behavior Provider
- Local Neural Net (DQN): Browser-trained reinforcement learning model with replay training.
- Local Rule Agent: Deterministic heuristics, no ML training required.
- Hybrid (DQN + LLM): DQN decides movement and action, local LLM handles richer language.
- Local LLM (WebLLM): In-browser model handles both decisions and chat.
- API Model: Uses your endpoint/model for decisions and chat, with local fallback on errors.
Model Settings
- Local LLM Model Key: WebLLM model identifier used for local language behavior.
- API Endpoint / Model / Key: Remote OpenAI-compatible inference target and credentials.
- Save Brain: Saves local DQN weights to IndexedDB.
- Reset Brain: Rebuilds a fresh local DQN model.
- Clear Memory: Clears world state and episodic memory in this browser.
Reward Tuning
Sliders scale reward components used by the learner. Higher values make that signal more influential during local training and adaptation.
Chat and Voice
- Chat box: Sends direct user messages; the familiar replies through the active provider or fallback rules.
- Auto-speak: Uses browser speech synthesis for replies and key familiar events.
- Activity log: Tracks user, familiar, system, and warning events with filter controls.