Free Lesson
High scale agentic memory with Turbopuffer
Part of RAG: Retrieval Augmented Gathering
60 min
Oct 1, 2026 1:00 PM
By continuing, you agree to Maven's Terms and Privacy Policy.
What you'll learn
How to iterate on agentic memory problems
How to debug agentic memory problems with a real dataset. Labeling evidence. Debugging why its not retrieved.
Take advanatge of scale to just "index everything"
How far can we get indexing everything into Turbopuffer, instead of modeling the perfect knowledge base?
Think in terms of layers of knowledge caches
Remembering evolving knowledge as a disposable cache, not a pristine beautiful organized platonic set of fats
Why this topic matters
Agents don't remember what they learn Jargon. Personal preferences. What works / doesn't work. In this talk, we'll explore one approach: just index all agentic traces into a high-scale vector database like Turbopuffer. Then reconstruct only the knowledge that we need, treating these neat and tidy bits of knowledge like a cache that expires, not facts that live forever. We'll discuss pros / cons of such an approach, and play with a fun solution.
