From Konshus
Independent, source-checked coverage of AI memory: assistant memory features, new memory companies, portability, privacy and research.
The AI memory news, once a week.
If you're interested in AI memory, want your own AI to remember you better, or just want to stay up to date, this is the newsletter for you. Every week we comb through the news for the big shifts in AI memory — new companies, new features from ChatGPT, Claude and Gemini, new research and new ideas — and sum up what matters in one short email.
For people who use AI every day, builders and founders, and anyone curious where AI memory is going.
2026-11-04 · Deep dive
Three new papers approach the same problem from different angles: what should an agent keep, when should it update what it knows, and how can anyone tell whether a memory helped? Their results are research findings, not changes to the assistants people use today.
2026-11-03
Claude now offers a way to import memories from other AI providers and export its own. New research shows why a portable memory also needs to be safe to retain, use and update.
2026-11-02 · Archive edition
New research examines three weak points in agent memory: acting on past interactions, using retrieved information, and revising conclusions when a remembered fact changes. Together, the papers offer a more demanding definition of what it means for an assistant to remember well.
2026-10-30 · Archive edition
California and New York are taking different approaches to chatbot memory involving minors. At the same time, new research is testing how agents can organize, check and consolidate what they remember.
2026-10-23 · Archive edition
MemTensor’s new models put an evolving memory inside the model’s computation. A separate study asks a useful counter-question: when is a dedicated memory system better than simply searching or supplying the conversation?
2026-10-19 · Deep dive · Archive edition
New research treats remembering as more than finding a relevant note. It asks whether a memory is sufficient, current and authorized to influence an answer.
2026-10-13 · Archive edition
A new study examines how repeated retrieval can make one old claim seem better supported than it is. The problem is not whether an assistant remembers, but whether it can tell an original source from its own echoes.
2026-10-08 · Archive edition
Google has announced a Gemini agent designed to carry context across work channels. Two new research papers show why building a useful memory is only part of the challenge: an agent must also know when a memory is appropriate to use.
2026-10-06 · Deep dive · Archive edition
A customer-service funding announcement and an exploratory research paper point to opposite sides of the same problem: useful context must be available when needed, without sitting in every agent’s active conversation.