TeleAI-UAGI/telemem

What it solves

TeleMem is an agent memory management layer designed to provide conversational AI with stable, long-term, and multimodal memory. It addresses common issues in AI agents, such as the inability to maintain consistent character settings across long interactions, high token costs, and the lack of a structured way to store and reason over video content.

How it works

TeleMem uses a pipeline of character-aware summarization, semantic clustering, and efficient storage to manage information. It extracts key facts from dialogues and uses LLMs to semantically merge similar memories to reduce conflicts. For text, it employs a dual-storage system using FAISS for fast vector search and JSON for human-readable metadata. For video, it implements a pipeline that extracts frames, generates captions, and stores them in a vector database, allowing agents to perform multi-step reasoning (ReAct-style) over video content.

Who it’s for

This project is intended for developers building multi-character virtual agent systems, long-memory AI assistants (such as companion AI or customer service), and agents requiring multimodal capabilities for video understanding and information retrieval.

Highlights

• Character-Specific Memory: Automatically creates isolated memory profiles for different characters to prevent persona confusion.
• Multimodal Video Support: A full pipeline for converting video to frames and captions for vector-based retrieval and reasoning.
• Mem0 Compatibility: Acts as a high-performance drop-in replacement for Mem0 with a compatible API.
• Local-First: Supports fully local execution using Qwen and FAISS, ensuring data privacy.
• High Performance: Optimized for faster write speeds (via batch flushing) and higher memory accuracy compared to baselines.