Blog
How esfis works underneath: context, retrieval, privacy and the decisions behind them. What shipped in each version is in the changelog.
- Agent usage, read straight off the disk The bar at the bottom of the Agents section shows how much each CLI has used and how the machine is doing, with no API and no key. What each agent records, what it does not, and why there are figures we would rather not show.
- The space first, the search second When two clients' context lives in the same place, one number has to be zero for ever: what crosses from one to the other. Why esfis does not filter after searching, how policies are applied and what is recorded about every delivery.
- When a rare word beats a vector To decide what reaches the agent, esfis combines BM25 with the macOS sentence vectors, and the vectors carry only a quarter of the weight. Why that is, how it picks a model when you write in Catalan, and the one question it still gets wrong.
- Who decides what gets remembered In esfis, what is worth remembering is chosen by the agent you are already talking to, not by a second model. The whole policy fits in the description of an MCP tool, and storing something costs nothing at all.
- Context, not memory Remembering across sessions is something the assistants already do. What none of them does is decide which context each AI should receive, where it came from and whether it still holds. So esfis is changing category.
- Introducing esfis A local memory, shared by every assistant you use. Why we built it, what it does today and what we have decided not to do.