[Feature] Data Editor: AI-Assisted Unused Column & Table Detection and Cleanup

[nocodeandy]

Introduce an AI-powered "Schema Cleanup" tool that identifies unused or likely-unused tables and columns across a Glide app and guides builders through safely removing them.

---

Builders working on mature or complex Glide apps frequently accumulate data schema debt — tables and columns that are no longer referenced by any component, workflow, computed column, or action.

The existing "Find Uses" / "Where Used" feature provides a starting point, but has known coverage gaps (e.g., Generate CSV actions, hidden tabs, API reads/writes, cross-app connections). As a result, builders cannot confidently use it alone to identify safe deletion candidates.

An AI-assisted cleanup workflow could analyze the full app schema, cross-reference known usage patterns, surface likely-unused fields, and walk builders through a guided, safe culling process — reducing bloat and improving app performance.

Current Problem

• "Find Uses" / "Where Used" has incomplete coverage, producing false negatives that erode builder trust in its safety signals.
• There is no bulk or systematic way to audit schema health across an entire app.
• Builders with hundreds of columns and tables have no practical path to identifying dead weight without manual, error-prone inspection.
• Unnecessary tables and columns add overhead to app load time, Data Editor performance, and builder cognitive load.
• Builders are left choosing between the risk of deleting something referenced (data loss/breakage) and the cost of keeping everything (sluggish apps, messy schemas).

Examples/Scenarios

• Mature production app: A builder who has iterated over months has accumulated 200+ columns across 15 tables. Many were part of deprecated workflows or prototype screens. There is no way to audit these systematically today.
• Post-migration cleanup: A builder who migrated from Google Sheets to Glide Tables brought over 40 columns they never ended up using. "Find Uses" shows zero uses, but they can't be sure if API integrations or cross-app references are silently consuming them.
• Performance investigation: A builder notices app load slowness and wants to identify and remove unused computed columns (which still evaluate at runtime) to reduce processing overhead.

Why This Matters

• Schema bloat is a real and common problem in production Glide apps; there is no current native solution beyond manual inspection.
• The known gaps in "Find Uses" mean builders cannot safely act on "zero uses" without additional confidence — AI analysis could fill this gap.
• Computed columns with zero uses still consume processing resources; cleaning them has direct app performance benefits.
• Providing a safe, guided cleanup workflow reduces the risk of accidental breakage, lowering the barrier to schema maintenance.
• This positions Glide more favorably for enterprise use cases where schema hygiene and app performance are operational requirements.

Suggested UX

• AI Schema Audit Panel
  • Accessible from the Data Editor or App Settings under a "Schema Health" or "Cleanup" section.
  • On-demand scan that analyzes all tables and columns against the full app graph (components, workflows, actions, visibility conditions, computed columns, API config, cross-app references).
  • Returns a categorized list:
    • 🟢 Likely safe to delete — zero detected uses across all known reference types
    • 🟡 Review recommended — low usage or only referenced in edge cases (e.g., hidden tabs, inactive workflows)
    • 🔴 In active use — confirmed references found; flagged for context only
• Guided Deletion Flow
  • Checkbox-select candidates for deletion, with inline "why we think this is unused" rationale per item.
  • Preview impact before confirming (e.g., "Deleting this column will also remove 1 computed column that depends on it").
  • Batch deletion with a single confirmation step.
  • Optional: export a "schema audit report" before any deletions are made.
• Coverage Transparency
  • AI tool clearly discloses which reference types were scanned vs. which remain opaque (e.g., external API reads not expressible in the builder).
  • "Confidence score" or similar signal per recommendation, so builders can calibrate their own risk tolerance.
    Notes:
• This feature is explicitly complementary to — not a replacement for — an improved "Find Uses" with full coverage. Both are needed.
• The AI layer adds value specifically in pattern recognition across large schemas and in surfacing "soft" signals (e.g., a column last used 6 months ago in a deprecated screen).

[nocodeandy]

Source: https://glideexpertscommunity.slack.com/archives/C05PM2YNJUF/p1776179240207509