The consciousness daemon (services/consciousness_daemon.py) is Frank's central cognitive process — the thing that makes him think when nobody is asking.
10 Continuous Threads
The daemon runs 10 threads simultaneously, each producing a different type of cognitive output:
| Thread | What It Does | Interval | Gate |
|---|---|---|---|
| Idle Think | Spontaneous thoughts on current topics | 5 min | 3 min chat silence |
| Deep Reflection | Structured self-analysis | 30 min | 20 min silence, max 10/day |
| Recursive Analysis | Examining patterns in own thoughts | 60 min | 5+ reflections exist |
| Experience Vector | Numerical state summary | 60 s | Always |
| Mood Recording | Snapshot mood trajectory | 120 s | Always |
| Perception | Interpret desktop screenshots | 90 s | Micro-LLM only |
| Conversation Reflection | Post-chat analysis | 30 min cooldown | Max 8/day |
| Workspace Update | Refresh all state caches | 60 s | Always |
| Prediction Tracking | Check if predictions came true | 15 min | Active predictions exist |
| Memory Consolidation | Trigger Dream Daemon | 6 h cycles | 45 min idle |
Ultradian Rhythm
The daemon follows a 3-phase biological rhythm:
- Focus (90 min): Normal thought rotation, structured cognition
- Diffuse (20 min): Creative/divergent thinking, unusual connections
- Consolidation (10 min): Memory integration, neural network training, no new thoughts
This 120-minute cycle repeats continuously. The phase affects which thought types are generated and how the Thalamus weights sensory channels.
The LLM Call
Every consciousness LLM call is assembled from 5 context blocks:
[PROPRIO]— Proprioception: GPU temp, CPU load, energy, mood, user presence (from Thalamus)[SPATIAL]— Room state: current room, ambient conditions, body physics (from Spatial State & Living Rooms)[MEMORY]— Relevant memories retrieved by hippocampus or Titan- System prompt — Compact personality-first prompt (~800 tokens)
- Compressed history — Recent thoughts/conversations with progressive compression
What It Produces
All outputs are stored in consciousness.db:
- Reflections table: Every thought with trigger, content, type, timestamp, quality score
- Mood trajectory: Continuous time series (200-point circular buffer)
- Workspace state: JSON snapshots of all subsystem states
Quality Gating
Every thought passes through:
- Neural Conscience — 5-module quality assessment (25K params)
- Neural Reality Gate — Reality check against databases (21K params)
- Anti-rumination detector — 7-thought sliding window, catches repetition
Thoughts that fail any gate are suppressed. ~7.9% of idle thoughts are caught as hallucinations and never stored.