Index Repository

Index your codebase for semantic code search.

Quick Start

cd review_eval
export OPENROUTER_API_KEY="..."
export QDRANT_URL="https://your-cluster.qdrant.io:6333"
export QDRANT_API_KEY="..."

uv run python -m review_eval.index_repo /path/to/repo --force

What Indexing Does

1. Scans the repository for code files
2. Chunks code by semantic boundaries (functions, classes)
3. Generates embeddings using OpenRouter's embedding model
4. Stores vectors in Qdrant for fast similarity search

Command Options

Basic Indexing

uv run python -m review_eval.index_repo /path/to/repo

Force Re-index

Delete existing vectors and re-index:

uv run python -m review_eval.index_repo /path/to/repo --force

Count Only

See how many chunks without indexing:

uv run python -m review_eval.index_repo /path/to/repo --count-only

Output:

Chunking repository: /path/to/repo
Found 5432 chunks

Environment Variables

Variable	Required	Description
OPENROUTER_API_KEY	Yes	For embedding generation
QDRANT_URL	Yes	Qdrant cluster URL
QDRANT_API_KEY	Yes	Qdrant authentication

Example Output

Indexing /path/to/repo...
  - Chunking Python files
  - Generating embeddings via OpenRouter (qwen/qwen3-embedding-8b)
  - Storing vectors in Qdrant (batched uploads)

Chunking repository: /path/to/repo
Found 10000 chunks, generating embeddings...
Uploading 10000 vectors to Qdrant...
Indexed 9938 chunks

Indexed 9938 chunks in 622.0s
Average: 16.0 chunks/sec

Incremental Updates

After code changes, update only changed files:

uv run python -m review_eval.update_embeddings /path/to/repo

This is faster than full re-indexing.

Indexing in CI/CD

GitHub Actions Workflow

name: Index Repository

on:
  workflow_dispatch:
  schedule:
    - cron: '0 0 * * 0'  # Weekly

jobs:
  index:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4

      - uses: actions/setup-python@v5
        with:
          python-version: '3.12'

      - name: Install dependencies
        run: |
          pip install uv
          cd review_eval && uv sync

      - name: Index repository
        env:
          OPENROUTER_API_KEY: ${{ secrets.OPENROUTER_API_KEY }}
          QDRANT_URL: ${{ secrets.QDRANT_URL }}
          QDRANT_API_KEY: ${{ secrets.QDRANT_API_KEY }}
        run: |
          cd review_eval
          uv run python -m review_eval.index_repo ${{ github.workspace }} --force

Supported File Types

Language	Extensions
Python	.py
TypeScript	.ts, .tsx
JavaScript	.js, .jsx
Rust	.rs
Go	.go
Java	.java

Performance Tips

Large Repositories

For repositories with 10,000+ files:

1. Start with a subset:

   # Index only src/
   uv run python -m review_eval.index_repo /path/to/repo/src
   

2. Exclude test files (they add noise):

3. Modify chunker to skip **/tests/**

4. Run overnight or on schedule

Optimizing Chunk Count

More chunks = more granular search but higher costs.

Default chunking creates chunks for: - Each function - Each class - Each method within classes

Troubleshooting

"Rate limit exceeded"

OpenRouter has rate limits. The indexer handles this with retries, but for large repos:

# Run in smaller batches
uv run python -m review_eval.index_repo /path/to/repo/src
uv run python -m review_eval.index_repo /path/to/repo/lib

"Connection refused" (Qdrant)

Verify your Qdrant URL:

curl -H "api-key: $QDRANT_API_KEY" "$QDRANT_URL/collections"

Missing files

Check the supported file types. Non-code files (markdown, config) are skipped.

Stale embeddings

Re-run with --force to refresh:

uv run python -m review_eval.index_repo /path/to/repo --force