SimpleMem is a Python-based memory framework for LLM agents that supports text and multimodal memories (image, audio, video) with an auto-routing mode and an optimization workflow using EvolveMem. Latest releases unify the package, add multimodal support, and document retrieval/config tuning.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
SimpleMem is an efficient lifelong memory system for LLM agents that stores, compresses, and retrieves memories with semantic lossless compression. It now supports multimodal memory (text, image, audio, video).
How it works
The workflow consists of storing dialogues or facts as structured memories, indexing memories using semantic embeddings and metadata, and retrieving memories by meaning. The default mode auto-detects the backend (text or multimodal) based on the first memory operation call. It provides an optimization path via an EvolveMem loop that evolves retrieval configurations (Evaluate → Diagnose → Propose → Guard).
Getting started
- Basic usage shows importing SimpleMem and creating an instance with auto mode:
from simplemem import SimpleMem
mem = SimpleMem() # mode="auto" — backend chosen by first call
- For multimodal usage, methods like add_text(), add_image(), add_audio(), add_video() are used depending on the first call.
- Advanced: optimize retrieval config with simplemem.optimize(mem, dev_questions, max_rounds=3) and save/load the config. Example:
config = simplemem.optimize(mem, dev_questions, max_rounds=3)
config.save("my_config.json")
config = load_config("my_config.json")
mem = SimpleMem(config=config)
- Parallel processing can be enabled during memory construction with create(..., enable_parallel_processing=True, max_parallel_workers=8, enable_parallel_retrieval=True, max_retrieval_workers=4).
Recent releases
- v0.3.0 — Unified SimpleMem package (2026-05-21): One import, auto-routing. Install in one step with
pip install -e .. - v0.2.0 — Omni-SimpleMem: Multimodal Memory (2026-04-03): Text, image, audio & video memory support.
- v0.1.0 — Initial Release (Text Only) (2026-03-10): Text memory framework based on semantic lossless compression.
Traction
- Stars: 3676
- Forks: 385
- Open issues: 3
Behind the repo
This repository emphasizes a unified package simplemem with auto-routing between text and multimodal backends. The project has an arXiv paper reference and multiple docs/folders for EvolveMem and Omni-SimpleMem.
Caveats
- License: MIT
- Created: 2026-01-01
- Last push: 2026-07-24
- Language: Python
- Requires Python 3.10+ and an OpenAI-compatible API key configured before memory construction or retrieval.