AI is useful in a starred library. It is also uniquely good at making confident mistakes.

A wrong auto-tag on one repository is annoying. A wrong auto-tag applied to two hundred repositories is a cleanup project. That is why Starcat treats AI as an advisor, not an author of record.

The rule

AI may suggest. Only you confirm writes.

In practice that means:

  • Tag recommendations appear as proposals you accept or dismiss
  • Summaries are generated for review, not silently overwritten into your notes
  • Nothing in your personal taxonomy changes because a model "felt sure"

This is slower than full automation. It is also how you keep trust in a tool you open every day.

Why confirmation is a feature

Your tags are a language. infra/observability and learn/later mean something specific to you. A model trained on the public web does not share that dialect.

Confirmation creates a feedback loop humans actually use:

  1. AI proposes a short list
  2. You keep what fits
  3. The library stays coherent

Without that step, the library slowly becomes someone else's ontology.

Conservative by design

Starcat unlocks AI behind Pro and BYOK-style provider settings. Even then, the product stance does not change: capability is gated; consent is not optional.

If you want a system that quietly rewrites your tags overnight, Starcat is the wrong app. If you want a second set of eyes that still asks before it writes — that is the point.