differentiation

vs input tools (Notion / Glasp / Readwise). They solve input — highlighter, note-taker, journal. They don't solve retrieval-by-others-in-your-voice. A Notion page is prose a visitor must read; StandMeet retrieval is dialogic — the visitor asks a specific question and gets a specific answer.

vs AI girlfriend / digital twin / personal LLM.

• AI companion optimizes your own emotional interaction with a bot (single-user, internal).
• Digital twin replicates you to act for a task (agency-replacement).
• Personal LLM trains a model on your writing (static, brittle, expensive).
• StandMeet is inter-personal access to articulated thought — the bot doesn't act as you or pretend to be you; it surfaces what you actually thought, when it's relevant to someone asking.

The nearest precedent isn't an AI product — it's a well-maintained personal site, but conversational.

The mechanical differentiators

The comparisons above are about positioning. These three are properties of the machine, and they are what a competitor would actually have to reproduce.

• Relevance is the owner's own links, not embedding similarity. No vector search, as a standing decision. corpus_links returns a note's outgoing links and backlinks; the agent decides whether to walk further. A RAG-shaped competitor ranks by proximity in a space nobody authored; here the edges were drawn by the person being asked about.
• A never-miss channel exists. corpus_grep is exhaustive literal/regex — if it is there, this finds it — which is an index property, not a similarity score. It is the mechanical backing for "prove what was actually said", and it is precisely what a vector store cannot promise. The eight retrieval tools are graded on this axis: search is ranked and can miss, grep cannot.
• It runs on the owner's own box. One instance per person, docker compose up, every table owner_id-scoped, no multi-tenant SaaS to scale. Sovereignty stops being a claim in the copy and becomes where the process is running (deployment).

The honest counterweight: none of the three is visible on the surface. A visitor cannot tell a never-miss channel from a good vector search by looking at one answer — the property only shows up in the case where the other kind fails, which is the case nobody demos.