---
title: Local-first is a feature, not a constraint
date: 2026-07-28
author: The Esment Team
description: Running memory on your own machine is not a limitation we accepted — it is the design that makes the product honest. Here is the architecture that makes it fast, private and portable.
tags: local-first, architecture, privacy
canonical: https://blog.esment.notas.ai/posts/local-first
---

# Local-first is a feature, not a constraint

*The Esment Team · 2026-07-28 · [HTML version](https://blog.esment.notas.ai/posts/local-first)*

There is a moment in every memory-product pitch where someone asks: "but
why not just put it in the cloud?" It is a fair question. The cloud is
easier to build, easier to sync, easier to sell. We chose local-first
anyway, and the reasons are not philosophical — they are practical.

## The privacy asymmetry

A memory store is the one dataset where the user's interests and the
company's incentives are structurally misaligned. The user wants the
assistant to know them; the company wants to know the user. Every cloud
memory product has to be evaluated under that tension. A local store
dissolves it: the data never leaves the machine, so there is nothing to
misuse, nothing to subpoena, nothing to leak.

## The architecture

The Esment engine is a single Rust binary that runs a SQLite-backed store
with:

- a **temporal knowledge graph** — entities, relations, and the memories
  that ground them;
- a **four-stage retrieval cascade** — FTS/BM25, HNSW vectors, graph BFS,
  and MMR reranking;
- a **commit DAG** — every mutation is a content-addressed, auditable
  commit;
- an **MCP server** — any assistant that speaks the Model Context
  Protocol can connect over stdio or HTTP.

All of it runs on the user's machine, answering in milliseconds. The
vector index is built locally; the LLM calls are the only thing that ever
touches a network, and only when the user configures a provider.

## The sync question

Local-first does not mean isolated. Esment syncs bidirectionally with the
hosted cloud when the user asks for it: additive in both directions,
conflict-safe, and with embeddings recomputed on each side. The cloud
becomes a mirror of your memory, never the owner of it.

## Why this wins

- **Speed.** No round-trip to a server for every recall; the cascade runs
  on-device in tens of milliseconds.
- **Portability.** The store is a file. Back it up, copy it, move it. It
  is yours in a way a database row can never be.
- **Honesty.** "Your memory stays on your machine" is a claim you can
  verify with `ls`. The enterprise edition ships fully local: no OpenAI,
  no Ollama, no network — a memory engine that runs on an air-gapped
  machine.

Local-first is often described as a constraint we accepted. It is not. It
is the feature that makes the whole product trustworthy enough to use.