How swagduck/mvdm works
MVDM is a deterministic, model-independent cognitive architecture with a bilingual local chat interface. The principal user workflow is entering a turn in the chat CLI, interpreting it into structured language/propositional representations, grounding and verifying a planned response against stored knowledge, then returning a grounded reply; withheld replies are logged rather than stored as self-utterances. Separate capabilities cover executive cognitive cycles, memory consolidation, identity, and autonomous development. Dashed links indicate runtime wiring not established by the supplied excerpts or README.
Main entry points: chat.py Chat CLI · loop.py Executive loop · ingest.py Utterance ingestion · engine.py Development engine
Components 23
Chat interface
- Chat CLI
mvdm/app/chat.py
Language processing
- Language runtime
mvdm/language/runtime.py - Utterance ingestion
mvdm/language/ingest.py - Unicode tokenizer
mvdm/language/tokenizer.py - Bilingual lexicon
mvdm/language/lexicon.py - Fact extraction
mvdm/language/extract.py - Response planning
mvdm/language/respond.py
Cognitive runtime
- Working workspace
mvdm/core/workspace.py - Deterministic baseline
mvdm/baseline/deterministic.py - Executive loop
mvdm/executive/loop.py - Action policy
mvdm/executive/policy.py - Value appraisal
mvdm/value/contracts.py
Knowledge and development
- Propositional frames
mvdm/reasoning/frames.py - Episodic memory
mvdm/memory/episodic.py - Memory consolidation
mvdm/memory/consolidation.py - Identity model
mvdm/identity/model.py - Belief graph
mvdm/reasoning/belief_graph.py - Development engine
mvdm/development/engine.py - Construction inducer
mvdm/development/constructions.py - Model registry
mvdm/development/registry.py
Persistence and trace
- SQLite WAL store
mvdm/storage/sqlite_store.py - Cognitive event bus
mvdm/core/events.py
Other components
- User
Connections 26
- User → Chat CLI: enters turn
- Chat CLI → Response planning: plans reply
mvdm/app/chat.py - Chat CLI → SQLite WAL store: verifies reply
mvdm/app/chat.py - Chat CLI → Belief graph: inspects beliefs
mvdm/app/chat.py - Chat CLI → Development engine: inspects development
mvdm/app/chat.py - Chat CLI → Model registry: requests rollback
mvdm/app/chat.py - Language runtime → Utterance ingestion: dispatches input
mvdm/language/runtime.py - Utterance ingestion → Unicode tokenizer: tokenizes text
mvdm/language/ingest.py - Utterance ingestion → Bilingual lexicon: uses vocabulary
mvdm/language/ingest.py - Utterance ingestion → Fact extraction: extracts meaning
mvdm/language/ingest.py - Fact extraction → Propositional frames: parses frames
mvdm/language/extract.py - Response planning → Propositional frames: uses structure
mvdm/language/respond.py - Response planning → Belief graph: grounds claims
mvdm/language/respond.py - Deterministic baseline → Value appraisal: appraises input
mvdm/baseline/deterministic.py - Executive loop → Working workspace: orchestrates cycle
mvdm/executive/loop.py - Executive loop → Episodic memory: retrieves episodes
mvdm/executive/loop.py - Executive loop → Action policy: requests decision
mvdm/executive/loop.py - Executive loop → Memory consolidation: runs consolidation
mvdm/executive/loop.py - Executive loop → Identity model: updates identity
mvdm/executive/loop.py - Episodic memory → SQLite WAL store: reads and writes
mvdm/memory/episodic.py - Identity model → Episodic memory: derives from episodes
mvdm/identity/model.py - Identity model → SQLite WAL store: reads entities
mvdm/identity/model.py - Development engine → Construction inducer: uses induction
mvdm/development/engine.py - Model registry → Development engine: reconstructs model
mvdm/development/registry.py - Development engine → SQLite WAL store: replays records
mvdm/development/engine.py - Chat CLI → Cognitive event bus: logs withheld reply
mvdm/app/chat.py