Expand description
Shodh-Memory Library
Edge-native AI memory system for autonomous agents. Optimized for deployment on resource-constrained devices.
§Key Features
- Tiered memory (working/session/long-term) based on cognitive science
- Local vector search (Vamana/DiskANN)
- Local embeddings (MiniLM-L6 via ONNX)
- Knowledge graph for entity relationships
§Edge Optimizations
- Lazy model loading (reduces startup RAM by ~200MB)
- Configurable thread count for power efficiency
- RocksDB embedded storage (no external database)
- Full offline operation
Re-exports§
pub use chrono;pub use parking_lot;pub use uuid;
Modules§
- ab_
testing - A/B Testing Infrastructure for Relevance Scoring
- auth
- backup
- P2: Backup & Restore System
- config
- Configuration management for Shodh-Memory
- constants
- Documented constants for the memory system
- decay
- Hybrid Decay Model (SHO-103)
- embeddings
- Embedding generation module
- errors
- Enterprise-grade error handling with structured error types and codes Provides detailed error information for debugging and client error handling
- graph_
memory - Graph Memory System - Inspired by Graphiti
- handlers
- HTTP API Handlers - Modular organization of the REST API
- integrations
- External integrations for syncing data sources to Shodh memory
- memory
- Memory System for LLM Context Management
- metrics
- Production-grade metrics with Prometheus
- middleware
- P1.3: HTTP request tracking middleware for observability
- mif
- Memory Interchange Format (MIF) v2
- query_
parsing - Modular Query Parsing System
- relevance
- Proactive Memory Surfacing (SHO-29)
- similarity
- Vector similarity search for semantic retrieval
- streaming
- Streaming Memory Ingestion for Implicit Learning
- tracing_
setup - P1.6: Distributed tracing with OpenTelemetry (OPTIONAL)
- validation
- Input validation for enterprise security Prevents injection attacks, ensures data integrity, protects against ReDoS
- vector_
db - Vector database module with pluggable index backends