The case for memory that your assistant cannot lose

We spent a year building Esment around a simple observation: the most expensive property of a good assistant is not the model — it is the context. The model is replaced every few months and gets better each time. The context is what you have told it, and it evaporates the moment the conversation ends.

The forgetting tax

Every assistant you use today starts from zero on every new chat. The consequences are not theoretical:

• you re-explain your preferences, your project layout, your constraints;
• the assistant repeats mistakes you already corrected;
• decisions made in one conversation are invisible in the next;
• and the assistant you trust with real work is the one that has known you the longest — which is precisely the one with no memory of you.

The industry answer has been "long context windows". But a larger window does not remember anything: it just lets you re-paste more. The window is a cache, not a memory.

What memory has to be

Once you accept that the assistant should persist what it learns, the design questions become sharp. Our answers:

Inspectable. If a system remembers things about you, you must be able to open it and see exactly what it knows. Esment stores every memory in a SQLite file on your machine, with a git-like commit history: every mutation is a signed, diffable commit. "What do you know about me?" is not a model's guess — it is a query.

Local. Your assistant's memory is a record of your work, your relationships, your decisions. Sending it to a cloud service you do not control turns a tool into a liability. The Esment store runs on-device; the cloud is an opt-in extension, never the default.

Deterministic. The deepest failure of assistant memory is not storage — it is recall. A model that only searches memory when it thinks to search will, on average, not think to search. So Esment injects a session brief into every conversation before the model answers: the people, projects and decisions that matter right now, retrieved by a cascade pipeline (full-text → vector → graph → rerank) in tens of milliseconds. Recall stops being the model's decision and becomes the harness's guarantee.

The compounding effect

Memory changes the economics of using an assistant. The first week, the difference is small. The third month, the assistant answers questions you have not asked it yet — because it knows the project, the constraints, the history. It stops being a chatbot and becomes a colleague with an excellent memory: annoying when it forgets, indispensable when it remembers.

That is the bet Esment is making: that the winning assistants of the next decade are not the ones with the biggest context windows, but the ones with the most reliable memory.