Somewhere Signals

Supermemory is moving upmarket toward B2B developer tooling. Their new messaging emphasizes API-first, not consumer. This is the same lane we're entering with the MI SDK. They have funding but no UMO equivalent — their memory is flat key-value, not semantic. Our structural advantage is the converter pipeline and receipted provenance.

Seeing a pattern: founders and operators are pasting Claude sessions into Notion, trying to manage their own second brain. The setup is manual, the maintenance is painful, and it dies within weeks. The insight: they want the output (searchable memory) but can't sustain the input (manual capture + curation). MI solves exactly this — live, self-cleaning, receipted. This is the demo moment.

The pgvector search was not filtering by user ID at the database layer — only at the application layer. Kasey patched all 4 layers: CTE materialization, application filter, search endpoint filter, and RLS policy. This is now production-safe for multi-user deployments. Key for the somewhere-house pilot: multiple team members can use the same MI instance safely.

MemPalace is building a beautiful consumer memory app. Their focus is personal knowledge management — not business teams, not API, not provenance. They have better UX than us today but no receipting, no multi-user isolation, no semantic pipeline. They'll own the "beautiful personal journal" space. We own the "business memory with receipts" space. Not a direct collision.

Decision made: MI will not build and maintain connectors (SharePoint, Notion, Google Drive). That is an integration maintenance burden we will never escape. Instead MI is the semantic filter between any input and any output. Customers bring their own connectors. We provide the UMO conversion, the receipted provenance, and the search. This is the right abstraction.