https://datatweets.com/courses/ai-agents/memory-and-state/Recent content in Memory and State on DATATWEETSHugoenCopyright (c) 2025 DatatweetsSun, 28 Jun 2026 09:00:00 +0200https://datatweets.com/courses/ai-agents/memory-and-state/lesson-1-the-messages-list-is-memory/Fri, 09 Jan 2026 09:00:00 +0200https://datatweets.com/courses/ai-agents/memory-and-state/lesson-1-the-messages-list-is-memory/An agent’s short-term memory isn’t a special feature — it’s the messages list the loop keeps appending to. Once you see that, the two memory problems become clear: the list grows past the context window, and it disappears when the session ends. This lesson frames both.https://datatweets.com/courses/ai-agents/memory-and-state/lesson-2-managing-context-truncation-and-budgets/Fri, 09 Jan 2026 09:00:00 +0200https://datatweets.com/courses/ai-agents/memory-and-state/lesson-2-managing-context-truncation-and-budgets/The simplest fix for a growing history is truncation: keep the first message and the most recent turns, drop the rest. Learn to truncate by whole turns, measure prompt size accurately with Claude’s count_tokens, and decide what to trim against a token budget.https://datatweets.com/courses/ai-agents/memory-and-state/lesson-3-summarization-memory/Fri, 09 Jan 2026 09:00:00 +0200https://datatweets.com/courses/ai-agents/memory-and-state/lesson-3-summarization-memory/Summarization keeps the gist of a long conversation while shrinking the messages list: fold the old turns into one summary message (written by a cheap model call) and keep the recent turns verbatim. Learn the compaction logic and when to summarize vs truncate.https://datatweets.com/courses/ai-agents/memory-and-state/lesson-4-long-term-memory-with-a-vector-store/Fri, 09 Jan 2026 09:00:00 +0200https://datatweets.com/courses/ai-agents/memory-and-state/lesson-4-long-term-memory-with-a-vector-store/Long-term memory stores durable facts as text notes and retrieves the relevant ones by similarity. Build a small VectorMemory with add and search, run it with a dependency-free keyword embedding, then swap in sentence-transformers or chromadb for real semantic search.https://datatweets.com/courses/ai-agents/memory-and-state/lesson-5-guided-project-give-atlas-memory/Fri, 09 Jan 2026 09:00:00 +0200https://datatweets.com/courses/ai-agents/memory-and-state/lesson-5-guided-project-give-atlas-memory/Put the module together: give Atlas short-term memory it manages with compaction, and long-term memory in a vector store it writes traveler preferences to and searches before planning — so it remembers and respects what it has learned.