A persistent, context-aware memory layer for AI agents and applications.
Large language models (LLMs) are stateless by nature. Every new prompt is processed in isolation — the model forgets what it said moments ago, what it learned in prior sessions, and what worked or failed before. This forces applications to re-inject context manually, results in fragmented and repetitive reasoning, and makes agents incapable of building on past experience.
The Context Memory Layer bridges this gap by giving AI agents a persistent, structured memory that is actively retrieved and injected at inference time. Instead of a single stateless prompt, agents get a curated slice of relevant context drawn from everything the system has ever seen or learned — just like how a human draws on working memory, experience, and learned facts to make decisions.
The memory layer is organized into four complementary memory types:
- Short-term: Current conversation context, working memory
- Long-term: Vector embeddings, knowledge bases, conversation history
- Episodic: Past interactions and outcomes
- Semantic: Facts, skills, world knowledge
These are complemented by retrieval mechanisms that select and surface the most relevant context for any given query.
- Continuity: Agents maintain coherent, informed behavior across sessions
- Contextual relevance: Retrieval surfaces the right knowledge at the right time
- Learning over time: Systems improve by accumulating and applying past outcomes
- Reduced redundancy: Relevant context is injected once, precisely, rather than dumped wholesale
The repo is organized into three learning modules, each with its own README.md:
RAG-Labs/
├── Agentic-RAG/ # Agentic RAG with tool use and multi-step reasoning
├── Graph-and-Vector/ # Hybrid graph + vector retrieval
├── Graph-RAG/ # Graph-based retrieval augmented generation
├── HybridRAG/ # Hybrid RAG combining sparse + dense retrieval
├── LLM-Wiki/ # LLM knowledge base with Wikipedia-style retrieval
├── MultiVector-RAG/ # Multi-vector retrieval (LangChain + ColBERT)
├── OCR-RAG/ # OCR-powered RAG for scanned documents
└── Vectorless-RAG/ # RAG without vector embeddings
MongoDB-Labs/
├── Lab 1 - Student Records Lookup/ # MongoDB Atlas CRUD, queries, indexing
├── Lab 2 - Student Enrollment Tracker/ # Full CRUD lifecycle, upserts
├── Lab 3 - Academic Performance Analytics/ # Aggregation pipelines, indexing
├── Lab 4 - Courses and Instructors/ # Schema design, $lookup, text search
├── Lab 5 - Enrollment Transactions/ # Transactions, change streams
├── Lab 6 - Scaling the University Database/ # Replica sets, RBAC, TLS
└── Lab 7 - AI Agent Memory Service/ # Documented capstone (build your own)
Audit-DB-Labs/
├── Lab 1 - Recording Agent Activity/ # Append-only audit logging on Supabase
├── Lab 2 - Modeling Runs, Spans, and Tool Calls/ # Hierarchical event modeling
├── Lab 3 - Querying Across the Hierarchy with JOINs/ # Recursive query patterns
├── Lab 4 - Filtering, Search, and Pagination/ # Search on audit history
├── Lab 5 - Metrics and Dashboards/ # Aggregate metrics and reporting
├── Lab 6 - Alerting on Anomalies/ # Alerting on unusual activity
├── Lab 7 - Enforcing Access Control and Detecting Tampering/ # RBAC, integrity
└── Lab 8 - Capstone Project - Agent Audit and Compliance Platform/ # Build it
Each module is standalone, with its own README.md covering setup and a full lab roadmap. Start with whichever module you want to learn — there's no required order between them.