A person stays in charge

RAG as corporate memory — how agents stop inventing identity

Large models are fluent. Fluency is not memory. If an agent writes a hex code, a legal notice, or a founder bio from “what sounds right,” you do not have a brand. You have a rumor with good grammar.

Retrieval-augmented generation (RAG) here means: look up house truth, then write. The corpus is small and ranked on purpose. It is not “the whole internet.”

Ranked corpora (highest first)

  1. Board-gated facts — `corporate-facts.json` (legal name, emails, what we will not claim).
  2. Policy filters — `forbidden-claims.json` and legal boilerplate. These reject sentences; they are not a writing style guide.
  3. Approved pages and posts — `status: approved` markdown that already survived review.
  4. Internal runbooks — how the host actually runs. Useful for operators; not scraped onto the public site.

If a fact is not in (1) or (3), the agent must not invent it. The correct output is a question or a `BLOCKED` note, not a plausible paragraph.

Chunking without theater

We chunk files the company already owns: kit YAML/JSON, approved markdown, runbooks. We do not scrape customer inboxes or live social threads into the public corpus. Chunks stay small enough to cite (a table, a section, a policy line) so a reviewer can see *which* file the sentence came from.

Ranking is boring on purpose: facts beat essays; approved pages beat drafts; nothing beats a Board unlock.

Generate-time filter

Retrieval is not enough. After a draft exists:

That stack does not promise platform outcomes. It makes invented identity expensive.

What this is not

MeltingFace