v0.3.0 — Unified SimpleMem package

SimpleMem v0.3.0 — Unified package

SimpleMem, Omni-SimpleMem, and EvolveMem now ship as a single simplemem package with auto-routing.

Highlights

• One import, auto-routing: from simplemem import SimpleMem. The first method you call selects the backend (text vs multimodal).
• Text memory (SimpleMem): semantic lossless compression, intent-aware retrieval.
• Multimodal (Omni-SimpleMem): text, image, audio, video via add_image / add_audio / add_video / query.
• Self-evolving retrieval (EvolveMem): simplemem.optimize(mem, dev_questions) tunes retrieval config offline.
• Packaging: pip install -e . installs the full stack; extras [server], [benchmark], [all]. Python 3.10+.

Notes

• All three paths (text / multimodal / optimize) verified end-to-end.
• MCP server and Docker remain text-only; multimodal and self-evolution over MCP are tracked in the README Roadmap.
• README and all 12 i18n translations updated to the unified API.

🤖 Generated with Claude Code

v0.2.0 — Omni-SimpleMem: Multimodal Memory

What's New

🧠 Omni-SimpleMem: Multimodal Memory

SimpleMem now supports text, image, audio & video memory.

• +411% on LoCoMo F1 over previous baselines
• +214% on Mem-Gallery F1
• 5.81 q/s retrieval throughput (3.5x faster)
• Built on three principles: Selective Ingestion, Progressive Retrieval, Knowledge Graph Augmentation

Other Changes

• README restructured with parallel SimpleMem (text) + Omni-SimpleMem (multimodal) organization
• Added Roadmap for upcoming multimodal infrastructure (cross-session, MCP, Docker)
• All 13 language READMEs updated
• Bug fixes and stability improvements

Full Changelog: v0.1.0...v0.2.0

v0.1.0 — Initial Release (Text Only)

SimpleMem v0.1.0 — Initial Release (Text Only)

An efficient memory framework based on semantic lossless compression for LLM agents.

Features

• Semantic Structured Compression — converts dialogues into compact, atomic memory units
• Online Semantic Synthesis — consolidates related memory fragments to eliminate redundancy
• Intent-Aware Retrieval Planning — dynamic search intent inference with parallel multi-view retrieval
• 43.24% F1 on LoCoMo benchmarks (26% improvement over Mem0)
• ~550 tokens per query — 30× fewer than full-context approaches
• MCP protocol support — compatible with Claude, Cursor, LM Studio, and more