Deep Memory

Deep-Memory: A Self-Evolving Knowledge Accumulation & Hybrid Retrieval System

English | 繁體中文
💡 Project note: This project is a deep refactor and rename of the original auto-skill project. We restructured it into a minimal, modular open-source skill pack, decoupled "code" from "data," and integrated local ChromaDB hybrid retrieval with BGE-Reranker re-ranking.

This skill turns your AI agent from a tool that forgets everything the moment a session ends into a self-evolving "second brain" that gets better the more you use it.

Deep-Memory is a meta-skill designed for AI assistants. It runs as a background knowledge system that automatically retrieves past experience during conversations, captures best practices, and — once a task succeeds — proactively writes the "successful approach" into your private knowledge base and indexes it, intelligently cutting down on token usage. You just keep working as usual; Deep-Memory runs automatically in the background.

Core Highlights

1. Genuinely Gets Stronger With Use
Traditional agents reset to zero the moment a conversation ends. Deep-Memory's Core Loop automatically checks a keyword index on every turn — if it recognizes a problem it has solved before, it directly recalls the "best solution" or "pitfall guide" from that time.

2. Cross-Skill Memory
Whenever you invoke another specific skill (coding, writing, drawing, etc.), Deep-Memory automatically checks its skill experience store.
Example: when you invoke remotion-video-gen, it proactively reminds you: "Last time we worked on this, setting FPS to 30 caused audio/video desync — we recommend switching to 60."

3. Proactive Experience Capture
You don't need to take notes by hand. When the AI detects that a task has been completed well, or you express satisfaction, it…