- Add daemon state tracking globals (mode, is_ruminating, cycles_by_type)
- Track trigger type and timestamp on each /cycle call
- Add DaemonStatusResponse model
- Add GET /daemon/status endpoint returning:
- mode: idle | conversation | autonomous
- is_ruminating: true when in rumination_free or corpus cycles
- last_trigger: type and timestamp
- cycles_breakdown: count by trigger type
- cycles_since_last_user: autonomous cycles since last user interaction
- time_since_last_user_seconds: elapsed time
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add /profile GET endpoint returning Ikario + David projections on 100 directions
- Compute Ikario state from Weaviate StateVector v1 + 113 thoughts
- Compute David tensor from user messages (SQLite) + declared profile
- Map direction categories to StateTensor dimensions via CATEGORY_TO_DIMENSION
- Calculate david_similarity as average cosine across 8 dimensions
- Result: 60.93% Ikario-David similarity (vs 100% when initialized from same source)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- api.py: REST API exposing LatentEngine via FastAPI
- POST /cycle: Execute semiotic cycle
- POST /translate: Translate state to language
- GET /state, /vigilance, /metrics, /health
- Loads embedding model and David profile at startup
- ~1.3s per cycle (embedding + dissonance + fixation)
- README2.md: Complete documentation of v2 architecture
- StateTensor 8x1024 explanation
- Module descriptions with code examples
- Amendments compliance
- Usage instructions
Start with: uvicorn ikario_processual.api:app --port 8100
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Implements the processual architecture based on Whitehead's Process
Philosophy and Peirce's Semiotics. Core paradigm: "L'espace latent
pense. Le LLM traduit." (The latent space thinks. The LLM translates.)
Phase 1-4: Core semiotic cycle
- StateTensor 8x1024 (8 Peircean dimensions)
- Dissonance computation with hard negatives
- Fixation via 4 Peircean methods (Tenacity, Authority, A Priori, Science)
- LatentEngine orchestrating the full cycle
Phase 5: StateToLanguage
- LLM as pure translator (zero-reasoning, T=0)
- Projection on interpretable directions
- Reasoning markers detection (Amendment #4)
Phase 6: Vigilance
- x_ref (David) as guard-rail, NOT attractor
- Drift detection per dimension and globally
- Alerts: ok, warning, critical
Phase 7: Autonomous Daemon
- Two modes: CONVERSATION (always verbalize), AUTONOMOUS (~1000 cycles/day)
- Amendment #5: 50% probability on unresolved impacts
- TriggerGenerator with weighted random selection
Phase 8: Integration & Metrics
- ProcessMetrics for daily/weekly reports
- Health status monitoring
- Integration tests validating all modules
297 tests passing, version 0.7.0
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Phase 3 - State Transformation:
- transform_state() function with alpha/beta parameters
- compute_adaptive_params() for dynamic transformation
- StateTransformer class for state management
Phase 4 - Occasion Logger:
- OccasionLog dataclass for structured logging
- OccasionLogger for JSON file storage
- Profile evolution tracking and statistics
Phase 5 - Occasion Manager:
- Full cycle: Prehension → Concrescence → Satisfaction
- Search integration (thoughts, library)
- State creation and logging orchestration
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- david_profile_declared.json: David's declared profile values from questionnaire
- scripts/embed_david.py: Python script to generate embeddings using BGE-M3 model
- questionnaire_david.md: Questionnaire template for profile values
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
