🧠
Agent State & Memory Management: Building Persistent AI Systems
Most AI agent tutorials build stateless chatbots. This course covers what production agents actually need: rolling-window memory with token budgeting, ChromaDB and Qdrant vector stores for long-term recall, LangGraph state machines for multi-turn workflows, Redis-backed checkpointing with replay strategies, and namespace-based multi-tenant isolation. Every module includes runnable Python sandbox exercises. You leave with five reusable components ready to drop into any LangGraph or Claude-based agent.
Duration
2 days
Level
Advanced
Price
9.99 EUR/month (all courses included)
Max group
12 participants
What you will learn
+Implement rolling-window conversation memory with token budgeting and Claude-powered summarization
+Build a vector-backed long-term memory store with ChromaDB and Qdrant for agent recall
+Design LangGraph state machines that handle multi-turn workflows with retry and branching logic
+Persist and replay agent state with Redis checkpointing and structured recovery strategies
+Isolate memory and state per tenant in a multi-user production system
Course program
Module 1: Conversation History Management — Rolling Windows, Summarization, and Token Budgeting
3h- Rolling-window memory manager with Claude token counting
- Claude-powered summarization pipeline: 85% history compression
- Hybrid strategy: recent messages + summarized older context
- Sandbox: build a token-budgeted conversation manager from scratch
Module 2: Vector Stores for Long-Term Agent Memory
3h- ChromaDB vs Qdrant for agent memory: when to use each
- Embedding strategy: what to store, how to chunk, when to forget
- Cosine similarity recall with metadata filtering
- Sandbox: long-term memory store with semantic search and GDPR-compliant deletion
Module 3: State Machines for Multi-Turn Agent Workflows
3h- LangGraph state graph: nodes, edges, conditional branching
- Retry and error recovery patterns without infinite loops
- Parallel tool execution with state merging
- Sandbox: multi-step research agent with approval gates
Module 4: Checkpoint and Recovery — Persisting Agent State
2h30- Redis-backed checkpointing with step-level granularity
- Replay strategies: resume from last checkpoint vs full replay
- TTL management and storage cost control
- Sandbox: fault-tolerant agent with automatic mid-run recovery
Module 5: Multi-Tenant State Isolation and User Context Management
2h30- Namespace isolation patterns: Redis key prefixing and ChromaDB collection-per-tenant
- Context injection: per-user preferences and memory at agent startup
- Audit trail: who changed what, when, and why
- Sandbox: multi-tenant customer support agent with full isolation
Ready to get started?
9.99 EUR/month — All courses included, cancel anytime