An all-in-one cloud workspace where a "note" is a database record with typed properties — title, status, tags, dates, relations to other records. Structure comes from relations, rollups, and multiple filtered views over the same underlying data, not from folders.
Best-in-class structured tagging — typed properties, cross-table relations, rollup aggregation, and task-specific views (Kanban, Calendar, Gallery) over one shared dataset. Mature collaboration, templates (Thomas Frank's Ultimate Brain, August Bradley's PPV), Notion AI, and a genuinely large existing user base.
Both reject "which folder?" in favour of tag/facet-driven retrieval, and both front their store with a serving-layer API rather than raw file access — the Notion API plays a similar automation role to Mimir's serving core, just without a lens engine behind it.
A compulsory governance layer Notion has no equivalent of — Notion's sharing permissions decide who can see a page, not whether a standard gets force-injected into a relevant answer. Mimir's classification tiers are enforced by the database itself (Postgres row-level security) rather than page-sharing settings, and its storage is an open Postgres schema rather than a proprietary block format — the research notes Notion's own export is documented as lossy: relations, rollups, and formula properties don't survive a migration out cleanly.
Notion's relational model — typed properties, rollups, and simultaneous filtered views of one dataset — is more mature and more usable today than Mimir's four blunt facets (tags, domain, classification, principal). Notion also has real multiplayer collaboration, a mature template ecosystem, and a far deeper automation surface (Zapier, Make, Readwise, the Notion API) than Mimir's six-endpoint daemon.
research/15-relational-database-first.md · research/_tools-and-stacks.md