Local-first software examples: where memory fits the pattern
By Matthew Firth, founder of ModelBrain · Sep 14, 2026 · 3 min readMost local-first software examples are documents or app state: your notes, your files, your board, living on your device first and syncing when convenient. Eight real ones below, and where an AI assistant's memory usually falls outside every one of them.
The pattern itself
Local-first, as Ink & Switch's original essay laid it out, means the primary copy of your data lives on your own device, network access is optional rather than required for the app to function, and no single vendor's server sits between you and your own files by necessity. It's a specific, checkable set of properties, not a marketing adjective.
Eight real examples
Notes are plain Markdown files in a folder you choose. The app reads and writes them directly; there's no server round-trip to open or save a note.
Same shape as Obsidian: a plain-text outliner over files on your disk, with sync as an optional add-on rather than a requirement to use it.
Your budget lives in a local file you own. The desktop app works fully offline; hosted sync between devices is optional.
A drawing saves to your device (or browser storage) by default. Real-time collaboration is a CRDT layer added on top, not a requirement underneath.
Same pattern as Excalidraw: canvas state lives locally first, and multiplayer sync is an add-on for the moments you actually need it.
Issue data is fully cached on your device and works offline; the app syncs in the background instead of gating every read on a network call.
An open-source project-management suite built local-first from the ground up, and self-hostable if you want the data on your own server instead of theirs.
A notes-and-objects workspace that keeps your data on-device and syncs peer-to-peer between your own devices rather than through a required central server.
For more, the community-maintained awesome-local-first list tracks a longer, growing set.
The gap: AI memory is usually the exception
None of the eight above is about an assistant's memory, and that's the point: ask most AI assistants, including ones built on top of otherwise local-first tools, to remember something, and what actually happens is a network call to a server that stores it in an account you don't control the files for. The local-first properties that app cared about for your notes quietly don't apply to what it remembers about you. That's not a contradiction on the tool's part; memory just hasn't usually been treated as data subject to the same guarantees as everything else.
Where ModelBrain fits
ModelBrain applies the same checkable guarantees (no account for local use, no query-time network call, a real file you can point at) specifically to an assistant's memory, rather than to notes or documents. What local-first software actually rules out is the detailed, checkable version of that claim, item by item, including the one deliberate exception (managed backup) rather than a blanket assertion.
What that looks like in pricing terms, since local-first and free aren't the same claim: the pricing page.