How theasthenosphere/samia works
SAM/IA is a local, biologically inspired persistent-memory system for AI agents and knowledge work. An agent invokes its MCP tools; recall combines episodic chainogram retrieval, fast hippocampal stores, and an optional semantic-fact arm into context, while write and consolidation paths preserve and refine memories. Autonomous maintenance runs decay and offline REM work, including fact extraction, contradiction handling, and integrity repair. The README documents these capabilities; sampled source confirms MCP delegation, retrieval behavior, and selected runtime wiring, so unsampled connections are kept conservative.
Main entry points: mcp_server_main.py MCP server · search.py Memory search · retrieval.py Chainogram retrieval · maintenanced.py Maintenance daemon
Components 21
Agent interface
- MCP server
samia/mcp_server_main.py - Memory search
samia/core/mcp_server/search.py - Memory writes
samia/core/mcp_server/write.py - Context tools
samia/core/mcp_server/context_tools.py
Recall and context
- Chainogram retrieval
samia/core/context_extension/retrieval.py - Semantic recall
samia/core/semantic_recall.py - Vector indexes
samia/core/vector.py - Hippocampal fast store
samia/core/hippocampus/__init__.py - Entity bridges
samia/core/entity_index.py
Memory lifecycle
- Tiered memory store
samia/core/tier.py - Episodic chains
samia/core/chain.py - Hebbian web
samia/core/bio/hebbian.py - Consolidation
samia/core/consolidation.py - Fact extraction
samia/core/fact_extractor.py - Integrity repair
samia/core/integrity/repair.py
Autonomous upkeep
- Maintenance daemon
samia/runtime/maintenanced.py - REM cycle
samia/runtime/rem_cycle/driver.py - Contradiction checks
samia/runtime/contradiction/passes.py - Local inference
samia/runtime/inference.py - Memory replay
samia/core/bio/replay.py
Other components
- AI agent
Connections 21
- AI agent → MCP server: invokes tools
- MCP server → Memory search: delegates search
- MCP server → Memory writes: delegates writes
- MCP server → Context tools: delegates context
- Memory search → Vector indexes: queries indexes
- Memory search → Hippocampal fast store: queries fast store
- Memory search → Hebbian web: reads coactivation
- Context tools → Chainogram retrieval: retrieves chains
- Context tools → Semantic recall: composes facts
- Chainogram retrieval → Episodic chains: reads chains
- Chainogram retrieval → Hebbian web: scores associations
- Chainogram retrieval → Entity bridges: queries bridges
- Memory writes → Tiered memory store: writes memories
- Consolidation → Tiered memory store: consolidates store
- Consolidation → Fact extraction: extracts facts
- Maintenance daemon → REM cycle: drives upkeep
- REM cycle → Consolidation: runs offline work
- REM cycle → Contradiction checks: sweeps candidates
- Contradiction checks → Local inference: requests judgment
- Maintenance daemon → Memory replay: runs replay
- Maintenance daemon → Integrity repair: runs repair