Brand voice

docs-keeper drafts read like a teammate, not generic AI prose. It does this by sampling your existing documentation on install and building a short style profile per repo.

What we sample

  • Your README.md (highest weight)
  • Any docs/**/*.md in scope
  • Your last 20 release notes from CHANGELOG.md

We extract: voice register (formal vs. casual), tense, person (you / we / it), heading capitalization, code-fence style, list markers, and approved terminology.

Where it's used

The profile is summarized into a single short paragraph that's prepended to the draft prompt. It influences word choice and structure — it does not change validation rules.

Default voice

If no docs exist yet (new repo), docs-keeper falls back to a safe default:

# Default brand voice

- Terse, friendly-formal, present-tense.
- Code-first examples.
- Sentence-case headings.
- No emoji.
- Use the second person ("you") when addressing the reader.

Pinning your own voice (Pro)

On the Pro tier, you can upload a custom voice profile in Settings → Brand Voice. It overrides the auto-sampled profile and persists across re-installs.

A good voice profile is 5–15 bullets that describe how to write, not what to write. Example:

# Acme Inc. — engineering voice

- Plain language. No jargon unless it's a term we coined.
- Show, don't tell — every claim has a code sample.
- Headings: Sentence case, no trailing punctuation.
- "We" for the company; "you" for the reader. Never "I".
- Always link to source files for type names and exports.
- Keep paragraphs under 4 sentences.