Store it. Restart. Recall it.
A local-first memory demo: add facts, restart the session, and watch them survive. Preset queries replay a trace recorded from the real SDK; arbitrary queries use a small browser scorer.
User prefers the Solarized Dark theme in VS Code and the terminal.
The team deploys on Fridays at 14:00 UTC, never after 16:00.
Deploys run through GitHub Actions and require a green CI run plus one approval.
Rollback is `kubectl rollout undo deployment/api` in the production cluster.
The platform team owns the API gateway and the deploy pipeline.
Priya leads the platform team and reviews production changes on Thursdays.
Project Atlas uses pnpm workspaces and a single SQLite file per service.
User is allergic to shellfish and avoids oyster sauce.
The staging database is reset every Monday at 03:00 UTC.
Incident 2026-10-14: the API gateway ran out of file descriptors after a config reload.
User prefers concise answers with code first and explanation second.
The design review happens on Wednesdays and needs a Figma link in the agenda.
Preset queries and packs are recorded from the real SDK. Additions and forgets made here persist in this browser's localStorage so a restart is visible; the real MemOS persists to SQLite at ~/.memos/memos.db. Arbitrary queries use a small lexical scorer in the browser, not the SDK's hybrid retrieval.
What this demo does and does not prove
- It proves the shape of the workflow: store, persist across a restart, rank, cite, and build a token-budgeted pack.
- The recorded queries and packs come from MemOS 1.6.27 running over the same twelve memories — including the trust and provenance fields shown on each row.
- It is not a SQLite-in-the-browser build. Additions live in localStorage; the SDK's real store is SQLite on your machine.
- Arbitrary queries are ranked by a deliberately small lexical model so the page stays fast and dependency-free. The real engine fuses keyword and embedding retrieval.