MXSTERMIND

mxstermind editorial

Semantic search turns your operation into memory

How mxstermind embeddings, pattern insights, and synthesis compound briefs, journal, deals, and role chat sessions into searchable institutional knowledge.

Published by mxstermind · 2026-06-15 · 8 min

Three months of data, zero recall

After ninety days on mxstermind you have dozens of briefs, hundreds of journal lines, deal notes, role chat transcripts, and content drafts. Without compounding, each session starts cold.

Sprint 10 adds pgvector embeddings (OpenAI text-embedding-3-small), a nightly cron to backfill, and real-time triggers on brief save, journal upsert, deal notes, and session end.

semantic_search() RPC returns cosine-ranked chunks scoped to your user_id. No cross-tenant leakage.

Where search shows up

Cmd+K: queries with three or more words route to semantic search; short queries stay keyword navigation. /app/search is the full UI with source type filters.

Finance role's search_knowledge tool calls the same index during role chat. RelatedInsights appears below the morning brief — Finance also found… with linked sources.

Pattern engine caches insights (deal velocity, brief timing, lead quality) with confidence scores. WeeklyPatternDigest on Monday Command. Analytics page shows PatternInsights card.

One console. Your entire operation.

Synthesis and role knowledge

Synthesis API builds a narrative strategic summary every six hours — exportable HTML, share with Finance, add to journal. Unlocks after 30 days of account age.

employee_knowledge table stores extracted expertise per Marketer, Finance, and Developer roles after sessions. Settings → Memory lets you edit, delete, and review what the platform learned.

Requires OPENAI_API_KEY on Vercel and migrations 047–050 in Supabase. Without them, the rest of mxstermind works — search and synthesis stay dormant.

Questions people ask AI about this topic

What is mxstermind semantic search?

Vector search over your briefs, journal, deals, role chat sessions, and more — via pgvector embeddings and /api/search/semantic.

What OpenAI model is used?

text-embedding-3-small for embeddings; chat models remain Gemini/Groq for AI Chat and briefs.

How do I enable semantic search in production?

Set OPENAI_API_KEY on Vercel, run migrations 047–050 in Supabase, enable the vector extension.

What is the synthesis page?

/app/synthesis — a strategic narrative view over your cached operation data, refreshable every six hours.

Start your Founder OS

Start free — morning brief, pipeline, and AI Chat in one console.