Skip to content

Changelog

All notable changes to Axon are documented here. This project follows Semantic Versioning.

[1.0.0-beta] - November 9, 2025

🎉 First Beta Release

This is the first beta release of Axon, representing the initial public version of the unified memory SDK for LLM applications.

✨ Added

Core Features

  • MemorySystem API - Main interface for store, recall, forget, compact, export operations
  • Multi-Tier Routing - Automatic routing across ephemeral, session, and persistent tiers
  • PolicyEngine - Orchestrates lifecycle decisions (promotion, demotion, eviction)
  • ScoringEngine - Importance scoring using access frequency, recency, and base importance
  • Router - Intelligent tier selection with fallback and capacity management

Storage Adapters

Adapter Use Case Status
InMemory Development/Testing ✅ Complete
Redis Ephemeral/Session ✅ Complete
ChromaDB Local Vector Storage ✅ Complete
Qdrant Production Vector DB ✅ Complete
Pinecone Managed Vector DB ✅ Complete

Embedders

  • OpenAI - text-embedding-3-small/large models
  • Voyage AI - Voyage embedding models
  • Sentence Transformers - Local open-source models
  • HuggingFace - HuggingFace inference API

Advanced Features

Audit Logging
  • Complete audit trail system with AuditLogger
  • Structured event logging with timestamps
  • Query/filter by operation, user, session, status
  • JSON export for compliance (GDPR, HIPAA)
  • Automatic rotation when max_events reached
Privacy & PII Detection
  • Automatic PII detection (emails, phones, SSNs, credit cards, IPs)
  • Auto-assigns privacy levels (PUBLIC, INTERNAL, SENSITIVE, RESTRICTED)
  • Detection results stored in entry metadata
  • User override capability
Structured Logging
  • JSON logging for production
  • Correlation ID tracking
  • Performance metrics via decorators
  • Environment-based configuration
Transaction Support (2PC)
  • Two-phase commit for atomic operations
  • Multiple isolation levels
  • Atomic multi-tier operations
  • Automatic rollback on failure
  • Context manager API
Compaction Strategies
  • Count-based compaction
  • Semantic similarity clustering
  • Importance-based selection
  • Time-based grouping
  • Hybrid strategies

Integrations

LangChain - AxonChatMemory implementing BaseChatMessageHistory - Backward compatibility with BaseMemory - AxonVectorStore for vector storage

LlamaIndex - AxonVectorStore implementing BasePydanticVectorStore - Query engine support - Metadata filtering

Configuration

  • MemoryConfig - Hierarchical configuration system
  • Policy Classes - Tier-specific policies
  • Pre-configured Templates:
  • DEVELOPMENT_CONFIG - In-memory, no embeddings
  • aggressive_caching() - Short TTLs, frequent compaction
  • balanced() - Recommended defaults
  • long_term_retention() - Extended TTLs, high capacity

Data Models

  • MemoryEntry - Pydantic-validated memory structure
  • Filter - Declarative query filtering
  • AuditEvent - Structured audit events
  • PIIDetectionResult - PII detection metadata
  • CompactionResult - Compaction operation results

🔄 Changed

  • Async-First API - All operations are async for better performance
  • Lazy Imports - Heavy dependencies loaded on-demand
  • Type Safety - Full type hints with mypy strict mode

🐛 Fixed

  • Memory export with missing embeddings
  • Compact dry-run OpenAI API key error
  • Missing await for async adapter access
  • Test expectations for async event ordering

📚 Documentation

  • MkDocs Site - Complete documentation with Material theme
  • Getting Started - Installation and quickstart
  • Core Concepts - Architecture deep dive
  • API Reference - Comprehensive API docs
  • Examples - 27 working code examples
  • Contributing - Development, testing, and style guides

✅ Testing

  • Overall Coverage: 97.8% passing (634/646 tests)
  • Unit Tests: 100/101 passing (99.0%)
  • Integration Tests: 92/94 passing (97.9%)
  • Framework Tests: 38/38 passing (100%)
  • Code Coverage: 42% overall, 70%+ on core modules

⚡ Performance

  • Single event loop design (Router)
  • Thread-safe adapter registry
  • Signature-based embedder caching

🔒 Security

  • Automatic PII detection
  • Four-tier privacy classification
  • Complete provenance tracking
  • Immutable audit trails

What's Working

✅ All core memory operations (store, recall, forget, compact, export, import)
✅ Multi-tier routing with automatic promotion/demotion
✅ 5 production storage adapters
✅ LangChain and LlamaIndex integrations
✅ Audit logging and PII detection
✅ Transaction support with 2PC
✅ Advanced compaction strategies

In Progress

🚧 Documentation optimization
🚧 Performance tuning
🚧 Security audit

Not Yet Implemented

❌ SQLite adapter
❌ CLI tools for backup/restore
❌ GraphQL API
❌ Real-time sync

Versioning Policy

Axon follows Semantic Versioning:

  • MAJOR: Incompatible API changes
  • MINOR: Backwards-compatible functionality
  • PATCH: Backwards-compatible bug fixes
  • Beta/RC: Pre-release versions

Future Releases

v1.0.0 (Stable) - Q1 2025

  • Complete documentation
  • Performance benchmarks
  • Security audit
  • Production deployment guide
  • SQLite adapter

v1.1.0 - Q2 2025

  • CLI backup/restore tools
  • Extended monitoring
  • Performance optimizations
  • Additional embedder integrations

v2.0.0 - Q3 2025

  • GraphQL API
  • Real-time sync
  • Multi-tenancy support
  • Advanced security features
  • PostgreSQL adapter with pgvector

Format: Based on Keep a Changelog
Versioning: Semantic Versioning