Knowledge Base / Second Brain

Fixes: the business’s knowledge lives in the owner’s head or scattered across drives, so nothing scales past them and onboarding is painful. Build a queryable company brain that AI (and staff) can pull from. Upsell tier: Knowledge Systems $3k+.

Stack

• Client-facing simple: Claude Project or Custom GPT with the company’s docs, SOPs, pricing, brand voice as context/knowledge files.
• Deeper / self-hosted: the “Company Brain” pattern (Nextcloud/OnlyOffice + pgvector + retrieval) for clients who need data control (see company-brain-offer). €150–400/mo hosted.
• Retrieval: hybrid keyword + vector search over their corpus.

field-tested build

1. Audit — what knowledge do people repeatedly ask the owner for? Where does it live now? What’s tribal vs. documented?
2. Optimize — this is mostly a content job: get the SOPs, pricing, FAQs, and brand voice written down and clean first. Garbage in, garbage out — the retrieval is only as good as the corpus.
3. Automate — load the clean corpus into a Claude Project / Custom GPT / vector store; set global instructions + brand voice; wire an access point (shared Claude project, internal chat, or Slack bot).
4. Verify — ask it the 10 questions staff ask the owner most; check every answer against ground truth (hallucination check). Fix the corpus where it’s wrong.

Time to ship

Moderate → Project — 1–2 sessions for a Claude Project; multi-week for a self-hosted brain.

Compliance

DPA/NDA before ingesting client data. If self-hosted with client data, data-handling terms are mandatory (company-brain-offer flags DPA-before-data).

Log in Notion

“Built company knowledge base ([N] docs). Answers [top questions] without pinging the owner. Onboarding time [before → after].” Track owner-interruption reduction.