Cleanup
Early accessFind the pages about things you removed or never built
Old articles keep describing features long after they are gone. Cleanup checks your pages against today's code and the history of your repos. It tells you which to keep, what to fix, and which to probably remove, with the proof. It never changes a page.
In early access: turned on per team. Write to hello@amendary.com to try it.
Works with
- Notion: works
- GitHub: works
- Zendesk: works
- Intercom: works
What it does
- You pick articles and pages, watched or not, from Notion, GitHub, Zendesk or Intercom, and the repos to check them against. Ticking a Notion page picks the pages under it too. Every connected repo is ticked by default.
- Each page gets one of two outcomes. Keep: the page still describes the product. Probably remove: what the page is about is gone from the code, or was never built.
- Each page also gets a one-sentence reason, how sure the check is, and 1 to 5 short guidelines on what to do with it: which statements to fix, what is missing, or what to check before you delete it.
- A contents page (only headings and links to its subpages) is skipped and costs nothing.
- You can export the report as CSV or Markdown for teammates who do not use Amendary.
- Run it before you start watching pages: watch the ones worth keeping, and delete the others yourself.
How it works
- The estimate comes first: a line for the pages and one for each repo. Nothing is read by a model until you start.
- Amendary reads each page and lists the names it uses for things in the code.
- Each repo is copied to a temporary folder on our server, one at a time, searched (the code today and every commit in its history), then deleted.
- Each page then gets its outcome from the evidence of every repo. Code checks the proof: a page is Probably remove only when what it is about is gone from every repo checked and every search finished. Without that proof, the page stays Keep.
- The report lists Probably remove first, then Keep, with links to GitHub at the commit.
What it never does
- It never edits, archives, deletes or maps a page on its own. The only actions are yours: Watch and map, or Mark done with what you did.
- It never marks a page Probably remove from a feature or label name, since a feature can live under another name in code. Only a name that reads as code (in backticks, a file path, a setting key, a URL), or an integration name with the commit that removed it, can prove something is gone.
- It never runs on a schedule and never adds items to the Queue.
Evidence and review
A Probably remove page links its proof: the commit that removed what it describes, or its absence from the product code history of every repo checked. Squashed or imported history can hide real code, so its guidelines say what to check before you delete it. When the content moved to another repo, a guideline says where.
A kept page with too little evidence is marked Not fully verified, and the reason says so. A dated record (changelog, decision record, incident review) or a page about process, policy or people is Keep.
A repo that could not be searched (too large, or could not be copied) blocks Probably remove, and the report says why. Pages that could not be read or that failed on our side are counted in the run’s status line. The report keeps references (repo, commit, file path, tag), a reason and guidelines per page, stored encrypted, never whole pages or files. What each outcome means is in the Cleanup help page.
Limits
- Up to 50 articles or pages per cleanup, from any destination. A translation is not checked on its own: it follows its source article.
- A repo over 300 MB is skipped, and the repos of one cleanup can add up to 1 GB. At most 400 names are searched per cleanup.
- Searching takes at most 20 minutes per repo and 90 minutes per cleanup. A name left unsearched cannot prove an absence.
- A cleanup that reaches your daily usage limit pauses and continues after the daily reset. If our server restarts, it picks up where it stopped. The owner or the person who started it can pause, resume or cancel it.
Pricing
A cleanup spends credits for reading the pages and giving each one its outcome. Contents pages are skipped and cost nothing. Searching code and history happens on our side and costs no credits. Credits are held when you start and the unused part is refunded at the end. Watching a page afterwards, and the daily checks that follow, cost nothing extra. See Billing & plans.
Where to go next
Try Amendary on your own help center
Connect Notion, GitHub, Zendesk or Intercom and pick the articles to watch. Your first check shows what Amendary would fix, with the change as proof.
Free for 14 days with no card, and every edit waits for your review. Or free forever with your own Anthropic API key. Code is never touched.