Re-Index Project Command

Force a complete re-index of the entire project, even if files haven't changed. Useful for troubleshooting, testing, or after schema changes.

Your Task

IMPORTANT: DO NOT DELETE THE DATABASE! This command re-indexes files by calling index_project with force=true. The database and existing records stay intact - only file embeddings are regenerated.

1. Get the current working directory:

   pwd
   

2. Warn user about the operation:

   ⚠️  Re-indexing entire project
   
   This will re-process all files, even if unchanged.
   For large projects, this may take time.
   
   Files in project: ~{FILE_COUNT}
   Estimated time: ~{ESTIMATE} seconds
   
   Continue?
   [Yes, re-index] [Cancel]
   

3. If user confirms, call index_project MCP tool:

   • Pass current working directory as project_path
   • Set force: true to re-index all files regardless of changes
   • DO NOT delete any files or databases
   • DO NOT use Bash commands to remove ~/.pixeltable/

4. Show progress and results:

   • Same format as /index-project command
   • Emphasize this is a full re-index
   • Report statistics

Example Interaction

User: /re-index

⚠️  Re-indexing entire project

This will re-process all 487 files in /Users/me/projects/api-service.
Estimated time: ~30 seconds.

Continue?

User: [Yes, re-index]

Re-indexing /Users/me/projects/api-service...

Progress: ████████████████████ 100% (487/487 files)

✅ Re-index complete!
- 487 files processed
- 487 files re-indexed
- 0 files failed
- Completed in 0.32s
- Storage: ~2.1 MB (refreshed)

All files have been re-indexed. Memory is fresh!

When to Use

Troubleshooting:

• Search results seem stale or incorrect
• Suspect index corruption
• After recovering from errors

Testing:

• Validating indexing performance
• Benchmarking on specific project
• Testing new indexing features

After System Changes:

• Pixeltable schema updated
• Embedding model changed
• Database migration completed

When NOT to Use:

• Normal workflow - use /index-project or auto-indexing
• After editing single file - use /index-file or let hook handle it
• Just to "be sure" - hash-based detection is reliable

Performance Expectations

Based on project size:

• Small (<100 files): <5 seconds
• Medium (100-500 files): 5-30 seconds
• Large (500-1000 files): 30-60 seconds
• Very Large (>1000 files): 1-2 minutes

Throughput: ~1,500 files/second on SSD with 8GB RAM

Comparison with /index-project

Feature	/index-project	/re-index
Skips unchanged files	✅ Yes (hash-based)	❌ No (forces re-index)
Use case	Normal indexing	Troubleshooting/testing
Speed	Fast (skips most)	Slower (processes all)
When to use	Regular workflow	Special situations

Alternative: Clear and Index

For a true "fresh start":

User: Can you clear the index and re-index from scratch?