← All articles in topic

Help center · Automation & AI

How draft suggestions work

On-device match against your directory. Server does not see full lists.

Last updated: 2026-05-242 min read

On this page

When a draft lands in the Extrack Automation inbox, the app runs on-device matching to suggest the most likely categories and people based on hints from the MCP client. Matches happen in your browser using fuzzy comparison against your own data; the MCP server never has your full category list or person directory. The matching utilities live in `resolveCategoryIdsFromNames` and `resolvePersonTagsForDirectory`.

Frequently asked

Why did the suggestion miss?

On-device matching is conservative — it only suggests when confidence is high. Ambiguous drafts show no suggestion; pick manually.

Can I improve suggestions?

They improve as your directory grows. Cleaner category and person names = better matches. Avoid duplicate-ish entries.

Related articles

Try this in Extrack

Open Extrack to follow the steps in a live workspace. Free for 30 days, no credit card.

Open Extrack