aiming-lab/SimpleMem
SimpleMem
SimpleMem is a persistent memory system for AI assistants designed to maintain context across different user chat sessions. It functions as a memory server and multimodal vector database that stores and retrieves information from text, images, audio, and video.
The project features a context compression engine that distills interaction histories into compact units to reduce token consumption. It utilizes a distributed memory orchestrator and worker-thread parallel processing to reduce latency when querying large-scale dialogue datasets.
The system implements a hybrid indexing approach combining semantic and keyword search for multimodal retrieval. It also includes a diagnostic framework for retrieval optimization that identifies failures and adjusts configurations to improve search precision.
Features
- Long-term Memory Stores - Provides a persistent memory store that allows AI assistants to maintain context across different user chat sessions.
- Multimodal Context Providers - Retrieves and assembles context from text, images, audio, and video to provide comprehensive situational awareness.
- Context Compression - Provides an engine for summarizing long interaction histories into compact units to minimize LLM token usage.
- Interaction Compression - Distills complex multimodal dialogue and media interactions into compact units to minimize LLM token consumption.
- Retrieval Optimization - Provides an iterative loop to diagnose retrieval failures and automatically tune configurations for better search precision.
- Multi-Modal Memory Stores - Operates as a persistent store managing text, images, and dialogue history across sessions using a standardized protocol.
- LLM Token Compression - Reduces token consumption by compressing interaction histories into compact, non-redundant units.
- Hybrid Vector-Keyword Indexing - Combines dense vector embeddings with inverted keyword indices to retrieve precise multimodal context.
- History Distillation - Compresses multimodal interaction histories into compact memory units to reduce token usage and eliminate redundancy.
- Multimodal Search - Indexes and retrieves information across text, image, audio, and video using a multimodal semantic search system.
- MCP Servers - Implements a Model Context Protocol server to expose persistent memory stores to AI assistants.
- Model Context Protocol Servers - Exposes a standardized Model Context Protocol interface for AI assistants to maintain persistent session context.
- Diagnostic Tuning - Diagnoses retrieval failures and automatically adjusts search configurations to improve memory recovery precision.
- RAG Optimizations - Includes a diagnostic framework to optimize the precision of memories recovered in RAG pipelines.
- Memory Processing Processors - Implements a memory processing pipeline that distributes retrieval and storage across multiple worker threads.
- Distributed Memory Orchestration - Coordinates multiple workers to reduce latency when querying large-scale dialogue datasets.
- High-Throughput Task Processing - Implements high-throughput task processing using distributed workers to accelerate memory building and retrieval.
- Parallel Retrieval Processing - Uses multiple background worker threads to parallelize memory building and query retrieval, reducing latency for large datasets.
