What if AI Actually Knew You?
Memory is a thought experiment brought to life. An open source framework for creating Personal Language Models that understand your context, remember your preferences, and evolve with you over time.
The Problem with AI Today
Every conversation with AI starts from zero. No memory. No context. No understanding of who you are.
Context Amnesia
LLMs forget everything after each session. You repeat the same context endlessly.
Generic Responses
AI treats everyone the same. No personalization based on your expertise or communication style.
Data Silos
Your digital life is fragmented. AI can't connect your emails, notes, code, and conversations.
The Personal Language Model
A new paradigm: AI that builds a persistent understanding of you from your digital footprint.
Three-Tier Memory
Short-term for recent context, long-term for semantic knowledge, persistent for core identity. Mirrors how human memory actually works.
Universal Ingestion
50+ data source adapters. Gmail, Slack, GitHub, Chrome history, local files, social media exports. Your complete digital footprint.
Privacy First
Everything runs locally. Your memories never leave your machine. Use local SLMs or enhance with cloud LLMs on your terms.
Continuous Learning
The PLM evolves with you. New conversations, corrected mistakes, and updated preferences are incorporated automatically.
Augmenting Intelligence
Memory doesn't replace LLMs. It makes them dramatically better by providing personal context.
Without Personal Context
With Personal Context
Built for Flexibility
Use powerful cloud LLMs for complex reasoning, or run entirely on local SLMs for privacy.
Hybrid by Design
Run Mistral 7B or Phi-3 locally for everyday queries with zero API costs. Seamlessly escalate to Claude or GPT-4 when you need frontier-level reasoning. Your context travels with you across providers.
Research Foundation
Memory synthesizes ideas from cognitive science, distributed systems, and modern ML research.
Titan Memory Architecture
Inspired by Google's Titan paper on neural long-term memory with surprise-based learning.
Learn more →Advanced RAG Patterns
Semantic search over personal knowledge graphs with multi-hop reasoning.
Learn more →Cognitive Architecture
Human-like memory consolidation: working memory, episodic memory, semantic memory.
Learn more →Model Compression
ZipLLM techniques for running personalized models on consumer hardware.
Learn more →This is a Research Project
Memory is my exploration of what's possible when AI has genuine personal context. It's open source, actively developed, and welcomes contributors who share the vision of AI that truly understands its users.