ALMA-memory v0.7.0

Added - Memory Wall Enhancements

• Memory Decay (alma/learning/decay.py)

  • Time-based confidence decay for aging memories
  • Configurable decay rates per memory type
  • Automatic refresh on memory access

• Memory Compression (alma/compression/)

  • LLM-powered memory summarization
  • Reduce storage footprint while preserving key insights
  • Configurable compression thresholds

• Retrieval Verification (alma/retrieval/verification.py)

  • Verify retrieved memories are still relevant
  • Cross-check with current context
  • Filter out stale or contradictory memories

• Retrieval Modes (alma/retrieval/modes.py)

  • standard: Default similarity-based retrieval
  • progressive: Start broad, narrow down
  • verified: Include relevance verification
  • budget: Token-aware retrieval with limits

• Trust-Integrated Retrieval (alma/retrieval/)

  • Agent trust scoring for memory weighting
  • Trust context propagation
  • Configurable trust thresholds

• Token Budget Management

  • Set maximum tokens for retrieved memories
  • Intelligent truncation preserving most relevant content
  • Budget allocation across memory types

• Progressive Disclosure

  • Start with summaries, expand on demand
  • Reduce context window usage
  • User-controlled detail levels

Changed

• Updated version badges in README
• Added quick comparison table vs Mem0 and Graphiti
• Comprehensive test suite added to repository (26,000+ lines)

Fixed

• Resolved ruff linting errors across codebase
• Fixed missing exports in retrieval and learning modules
• ESLint errors resolved in JavaScript SDK

---