Index Command

CLI reference for the index command

Index Command

Bulk index GitHub issues and pull requests to the vector database for semantic search.

Syntax

simili index [OPTIONS]

Options

Option	Short	Type	Description	Default
--repo	-r	string	Repository: owner/name	Required
--config	-c	file	Configuration file	.github/simili.yaml
--since	-s	string	Index issues since (ISO date or issue number)	-
--limit		number	Maximum issues to index	unlimited
--workers	-w	number	Parallel workers	5
--include-prs		bool	Include pull requests	true
--token		string	GitHub token (falls back to GITHUB_TOKEN)	-
--dry-run		bool	Simulate without writing	false
--help	-h	bool	Show help message	-

Examples

Index last 30 days

simili index --repo owner/repo --since 2026-01-01

Index since specific date

simili index --repo owner/repo --since 2024-01-01

Limit results

Index at most 200 issues:

simili index --repo owner/repo --limit 200

Parallel processing

Use more workers for faster indexing:

simili index --repo owner/repo --workers 10

Index everything (issues only)

simili index --repo owner/repo --since 2020-01-01 --include-prs=false

Index with PR collection

When qdrant.pr_collection is set in config, PRs are routed to a dedicated collection:

simili index --repo owner/repo --config .github/simili.yaml

Output

Fetching issues from owner/repo...
  Found 523 issues

Creating Qdrant collection...
  ✓ Collection created

Indexing with 5 workers...
  [====>----] 45% (235/523) - Speed: 12 issues/sec

Date format

ISO 8601 dates

• 2024-01-01
• 2024-01-01T12:00:00Z

Performance

Typical Speed:

• With 1 worker: ~2-3 issues/second
• With 5 workers: ~10-15 issues/second
• With 10 workers: ~20-25 issues/second

Depends on:

• Issue complexity and comment count
• API response times
• Network latency

Process

For each issue:

1. Fetch issue details from GitHub
2. Fetch all comments
3. Combine title + body + comments
4. Split into chunks (recursive character splitter)
5. Generate embeddings for each chunk
6. Upsert to Qdrant with metadata

For PRs (when --include-prs is enabled):

1. Fetch PR details and changed file paths
2. Embed: Title: ...\n\nBody: ...\n\nChanged Files:\n- path/a
3. Route to pr_collection if configured, otherwise to main collection

Configuration

Requires a valid configuration with:

• Qdrant connection details
• Gemini or OpenAI API key
• Target repository

See Configuration Overview.

Tips

1. Start with a small batch

   simili index --repo owner/repo --since 2026-01-01 --limit 50
   

2. Use multiple workers for large repos

   simili index --repo owner/repo --workers 10
   

3. Run during off-hours to avoid API rate limits

4. Set a PR collection to separate PR and issue search results

Troubleshooting

Rate limited

Error: 429 Too Many Requests

Reduce worker count:

simili index --repo owner/repo --workers 2

Out of memory

Reduce workers or use --limit to process in batches.

Collection already exists

Re-indexing is safe — upsert updates existing vectors.

Next steps

Process command

Process individual issues

PR duplicate command

Detect duplicate pull requests