SOP-009: Codebase Indexing

Document Control

Field	Value
SOP ID	SOP-009
Version	1.0
Status	Active
Last Updated	2024-12

Purpose

This SOP covers Cursor's codebase indexing feature, which enables semantic search across your workspace for AI-powered code discovery.

Scope

• How indexing works
• Privacy and security
• Configuration options
• Troubleshooting indexing issues

Prerequisites

• Cursor installed and configured
• Workspace open with code files

---

Procedure

How Indexing Works

Indexing Pipeline:

1. Sync - Files securely synced to Cursor servers
2. Chunk - Code split into meaningful chunks (functions, classes, logical blocks)
3. Embed - AI converts chunks to vectors (semantic meaning captured)
4. Store - Vectors stored in vector database
5. Query - Your search → vector → similarity match
6. Retrieve - Most similar code returned with context

Step 1: Check Indexing Status

View Status:

1. Open Cursor Settings: Cmd+Shift+J / Ctrl+Shift+J
2. Navigate to Indexing & Docs
3. See indexing progress percentage

Semantic Search Available

Semantic search becomes available at 80% completion. You can start using it before full indexing completes.

Step 2: Understand What Gets Indexed

File Type	Action
New files	Automatically added
Modified files	Re-indexed with new embeddings
Deleted files	Removed from index
Large/complex files	May be skipped for performance
Ignored files	Excluded (see below)

Automatic Sync:

• Checks every 5 minutes
• Only changed files re-processed
• Minimal performance impact

Step 3: View Indexed Files

1. Go to Cursor Settings → Indexing & Docs
2. Click View included files
3. Opens .txt file listing all indexed paths

Step 4: Configure Ignore Files

Files matching ignore patterns are excluded from indexing.

Ignore Sources:

• .gitignore - Respected by default
• .cursorignore - Cursor-specific ignores

Create .cursorignore:

# Ignore large data files
*.csv
*.parquet
data/

# Ignore generated code
dist/
build/
node_modules/

# Ignore secrets
.env*
secrets/

Improve Accuracy

Ignoring large non-code content files improves search accuracy by reducing noise.

Step 5: Force Re-Index

If indexing seems stale or corrupted:

1. Open Command Palette: Cmd+Shift+P / Ctrl+Shift+P
2. Search: "Cursor: Resync Index"
3. Wait for re-indexing to complete

---

Privacy & Security

File Paths:

• Encrypted before transmission
• Project structure remains confidential

Code Content:

• Never stored in plaintext on servers
• Only held in memory during processing
• Discarded after embedding generation

Embeddings:

• Stored for search functionality
• Cannot be reversed to original code

---

Semantic Search vs Grep

Type	Search: "navigation"	Result
Grep	Pattern matching	Finds files containing "navigation" literally. Misses header.tsx (has nav menu, not the word)
Semantic	Meaning-based	Finds header.tsx, sidebar.tsx, nav-menu.tsx - understands they're related

Cursor uses both:

• Grep for exact patterns
• Semantic for conceptual matches
• Combined for best results

Benefits of Semantic Search:

• Faster results - Compute happens at index time, not search time
• Better accuracy - Custom-trained models for code
• Fewer clarifications - Find what you mean, not just what you type
• Conceptual matching - Find code by what it does

---

Configuration

Navigate to Cursor Settings → Indexing & Docs

Setting	Description
Auto-index new repos	Automatically index when opening new projects
View included files	See list of indexed file paths
Show Settings	Configure ignore patterns

---

Frequently Asked Questions

Where can I see all indexed codebases? Cursor Settings → Indexing & Docs shows current workspace. Historical indexes are managed automatically.

How do I delete all indexed codebases? Indexes are workspace-specific. Close workspace and delete .cursor folder to remove local cache.

How long are indexes retained? Indexes persist as long as you actively use the workspace. Unused indexes are cleaned up automatically.

Does Cursor store my source code? No. Code is processed in memory for embedding generation then discarded. Only vectors are stored.

How does team sharing work? Each team member has their own index. Indexes are not shared between users.

Multi-root workspaces? Supported. Each root folder is indexed separately.

---

Verification Checklist

• [ ] Indexing status shows percentage in settings
• [ ] Can view list of included files
• [ ] .cursorignore file is respected
• [ ] Semantic search returns relevant results
• [ ] Large binary files are excluded

Troubleshooting

Issue	Solution
Indexing stuck	Check network, try resync
Files not indexed	Check ignore patterns
Wrong results	Re-index, check file types
Slow indexing	Large codebase normal, ignore non-code
Privacy concerns	Enable Privacy Mode in settings

See Also

• SOP-010: Project Rules
• SOP-011: @ Mentions
• Troubleshooting Guide