How Changelog Generation Works

ShipRelay uses Claude (by Anthropic) to transform raw commits and PRs into user-friendly changelogs.

The pipeline

  1. Fetch — ShipRelay retrieves all commits and merged PRs between two version tags via the GitHub API
  2. Analyze — Commit messages and PR descriptions are sent to Claude for analysis
  3. Group — Claude categorizes changes into three sections: Features (new capabilities), Fixes (bug fixes), and Improvements (enhancements, refactors, performance)
  4. Draft — The structured output is saved as an editable draft in your dashboard

Conventional commit support

ShipRelay recognizes conventional commit prefixes like feat:, fix:, perf:, chore:, docs:, refactor:, test:, build:, ci:, and style:.

  • When commits follow this format, ShipRelay uses prefixes as a strong signal for better grouping accuracy.
  • When commits do not follow conventional format, ShipRelay falls back to AI judgment-based categorization.
  • Both paths produce high-quality output — conventional commits simply give the model a better head start.
  • Infrastructure-oriented commits such as chore: and ci: can be marked with exclude_suggested for user-facing exclusion.

What data is sent to the AI

  • Commit messages (the text, not the code diff)
  • PR titles and descriptions
  • The repository name and version tags (for context)

What is NOT sent

  • Source code or file contents
  • Your personal information
  • Any data from other repositories

Editing drafts

Drafts open in a rich-text editor (Tiptap). You can freely edit text, reorder items, add or remove entries, and change formatting. The editor auto-saves your changes.

Regenerating

If you're not happy with the output, click Regenerate. You'll see a confirmation dialog (regeneration replaces the current draft). You can also switch the audience tone before regenerating — the AI will produce a substantively different output.

Generation limits

  • Free tier: 3 changelogs per month
  • Solo and Team: Unlimited changelogs
  • Rate limit: 10 generations per user per hour (to prevent abuse)

Tip

The quality of the AI output depends on the quality of your commit messages and PR descriptions. Teams that write descriptive PR titles get noticeably better changelogs.
Previous
Connecting GitHub
Next
Audience Tones