Blog
Notes on memory, retrieval, and what it takes to make an assistant actually remember you. Shipping notes live in the changelog.
- Introducing Esfis One local memory for every assistant you use. Why we built it, what it does today, and what we are deliberately not doing.
- Cascade retrieval, or why we run four searches instead of one Vector search alone loses exact names and dates. Keyword search alone loses paraphrase. We run both, expand through the graph, and rerank. Here is what each stage buys.
- Memory you can diff Every change to your memory is a commit. Here is what that makes possible, from "what did I know in June" to undoing a bad import.
- One message in, four memories out What extraction actually does to a paragraph, and why we split aggressively instead of storing whole messages.
Subscribe
One email a month when something worth reading goes up. Or take the RSS feed.
No tracking pixels, no sharing.